MODELING OF PERIODIC DIELECTRIC STRUCTURES ... · neous dielectric regions forspecificusein planar...
Transcript of MODELING OF PERIODIC DIELECTRIC STRUCTURES ... · neous dielectric regions forspecificusein planar...
MODELING OF PERIODIC DIELECTRIC STRUCTURES
(ELECTROMAGNETIC CRYSTALS)
by
John David Shumpert
A dissertation submitted in partial fulfillmentof the requirements for the degree of
Doctor of Philosophy(Electrical Engineering)
in The University of Michigan2001
Doctoral Committee:Professor Linda P. B. Katehi, ChairAssociate Professor Kamal SarabandiEmeritus Professor Thomas B. A. SeniorProfessor Wayne E. Stark
“Almighty God, Who hast created man in Thine own Image, and made him a
living soul that he might seek after Thee, and have dominion over Thy crea-
tures, teach us to study the works of Thy hands, that we may subdue the earth
to our use, and strengthen our reason for Thy service; and so to receive Thy
blessed word, that we may believe on Him whom Thou hast sent, to give us the
knowledge of salvation and the remission of our sins. All which we ask in the
name of the same Jesus Christ, our Lord.”
— James Clerk Maxwell (1831–1879)
c© John David Shumpert 2001All Rights Reserved
Sola Deo Gloria
ii
ACKNOWLEDGMENTS
I am grateful to Professor Linda P. B. Katehi for serving as my dissertation chair and for
her continued support of my work. I also appreciate Professors Kamal Sarabandi, Thomas
B. A. Senior, and Wayne E. Stark for their welcomed comments and suggestions and for
graciously giving of their time to serve on my dissertation committee.
I am grateful to the National Science Foundation without whose financial support, in
the form of the National Science Foundation Graduate Research Fellowship, some of my
graduate work would not have been possible. I also gratefully acknowledge the support of
the Department of Defense Research and Engineering (DDR&E) Multidisciplinary Univer-
sity Research Initiative (MURI) on “Low Energy Electronics Design for Mobile Platforms”
managed by the Army Research Office (ARO) under Grant DAAH04–96–1–0377.
The completion of an extended period of directed research that is required to obtain
an advanced degree is the product of hard work, perseverance, and the continued support
of family, friends, and my fellow comrades-in-arms. The Radiation Laboratory at the Uni-
versity of Michigan is not simply the group of offices, laboratories, and classrooms that
comprise the major portion of the third floor of the EECS Building; it is the collection
of faculty, staff, and students who live and work together every day. In particular, staff
members Mike Angelini, Michelle Shepherd, Carol Truskowski, and Patti Wolfe were vaults
of hidden knowledge for myself and the other graduate students in the Lab. Carol, your
indefatigable spirit is an invaluable asset to the graduate students in the Lab. I would like
to thank my comrades-in-arms, current and former University of Michigan Radiation Lab
graduate students Lars Andersen, Scott Barker, Andy Brown, Mark Casciato, Bill Chappell,
Rashaunda Henderson, Katherine Herrick, Stephane Legault, Sergio Pacheco, Ron Reano,
and Kyoung Yang, for the hours of enjoyable conversation and fun. I am thankful to Bill
Chappell, Saqib Jalil, and Tom Ellis, for their help with the fabrication and design aspects
of the slot antenna work.
iii
I would like to especially thank Mark Casciato, Katherine Herrick, Stephane Legault,
and Sergio Pacheco, without whom my short time at the University of Michigan would
have seemed an eternity. I will dearly miss our many “not-so-technical” discussions. Your
friendship has been and remains an invaluable asset to me and to my family.
I am very blessed to have the love and support of my family. Dad, I am now your
colleague; you have always been my biggest fan. Mom, I thank you for your ever encouraging
and joyful spirit. Simple words seem inadequate to express my love for you both and my
appreciation for your continual prayers throughout this journey. Jim and Stephanie, your
confidence in my ability to persevere was a welcomed voice at the other end of the telephone.
Caleb, I thank you for your unconditional love for Daddy.
Breanna, to you I express my deepest appreciation. Your prayers, unconditional
love, constant encouragement, and unwavering devotion sustained me during the
many long days of deriving pages of equations or sitting alone in front of a com-
puter screen desperately debugging long Fortran codes. May our Lord, Jehovah,
the God of the Bible, bless you with the desires of your heart for providing me
with the strength and perseverance to complete this work.
iv
PREFACE
Electromagnetic band-gap (EBG) materials, often termed electromagnetic crystals or
photonic band-gap (PBG) materials, are found to have unique properties that are advanta-
geous for applications involving semiconductor integrated circuits (ICs). Preliminary results
suggest that at microwave and millimeter-wave frequencies the propagation characteristics
of these materials can be manipulated by carefully designing and fabricating periodic struc-
tures composed of regions of differing dielectric constants. Researchers from the diverse
fields of classical electromagnetics, solid-state physics, optics, material science, condensed
matter physics, and semiconductor physics, are actively contributing to this rediscovered
field of physics. Active areas of electromagnetic and photonic crystal research include but are
not limited to microwave and millimeter-wave antenna structures, quasi-optical microwave
arrays, photonic crystal integrated circuits, high- and low-Q electromagnetic resonators,
quantum optical electromagnetic cavity effects, and optical nano-cavities. In addition, sonic
band-gap materials or artificial acoustic crystal substrates, are being developed and could
impact sonar.
Recent attention in the field of EBGs has focused on the elimination of surface-wave
formation in planar microwave and millimeter-wave antenna applications. A number of
numerical techniques, including finite difference methods, finite element methods, spectral
domain analysis, and integral equation (IE)/moment methods (MoM), have been imple-
mented to help understand the properties of surface and leaky waves on a layered, periodic
structure. The IE/MoM approach is a useful approach because information about the
physical system (propagation constants, mode structure, etc.) can be obtained easily and
directly from the formulation. A full-wave IE/MoM code has been developed to determine
the band structure (propagating modes) of periodic one- and two-dimensional inhomoge-
neous dielectric regions for specific use in planar antenna applications. This work derives
the integral equation approach along with several examples showing the unique spectral
characteristics of EBGs.
v
TABLE OF CONTENTS
DEDICATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
PREFACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
LIST OF APPENDICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
CHAPTER
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 One-Dimensional Periodic Dielectric Structures . . . . . . . . . . . 102.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.2 Fundamental Concepts . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . 122.2.2 Propagation of Waves in a Periodic Medium . . . . . . . . . 13
2.3 Plane Wave Expansion Method . . . . . . . . . . . . . . . . . . . . 142.3.1 Analytical Techniques . . . . . . . . . . . . . . . . . . . . . 142.3.2 Matrix Solution of the Eigensystem . . . . . . . . . . . . . . 18
2.4 Formulation of Integral Equations . . . . . . . . . . . . . . . . . . . 232.4.1 Derivation of Electric Field Integral Equation . . . . . . . . 232.4.2 Equivalent Polarization Currents . . . . . . . . . . . . . . . 242.4.3 Method of Moments Formulation . . . . . . . . . . . . . . . 282.4.4 Matrix Solution . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Experimental Verification . . . . . . . . . . . . . . . . . . . . . . . 392.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3 Two-Dimensional Periodic Dielectric Structures . . . . . . . . . . . 453.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.2 Plane Wave Expansion Method . . . . . . . . . . . . . . . . . . . . 47
3.2.1 Analytical Techniques . . . . . . . . . . . . . . . . . . . . . 473.2.2 Matrix Solution of the Eigensystem . . . . . . . . . . . . . . 53
3.3 Formulation of Integral Equations . . . . . . . . . . . . . . . . . . . 56
vi
3.3.1 Derivation of Electric Field Integral Equation . . . . . . . . 563.3.2 Method of Moments Formulation . . . . . . . . . . . . . . . 613.3.3 Matrix Solution . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4 Effective Medium Theory (EMT) . . . . . . . . . . . . . . . . . . . 673.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4 Scattering From an Inhomogeneous Doubly Periodic DielectricLayer Above a Layered Medium . . . . . . . . . . . . . . . . . . . . . 73
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.2 Formulation of Integral Equations . . . . . . . . . . . . . . . . . . . 74
4.2.1 Derivation of Electric Field Integral Equation . . . . . . . . 744.2.2 Integral Equations . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Numerical Implementation . . . . . . . . . . . . . . . . . . . . . . . 824.3.1 Impedance Matrices . . . . . . . . . . . . . . . . . . . . . . 834.3.2 Excitation Matrices . . . . . . . . . . . . . . . . . . . . . . . 91
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 914.5 Series Acceleration Techniques for the Periodic, Free-Space Green’s
Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.5.1 Acceleration Techniques Applied in This Work . . . . . . . 954.5.2 Acceleration Techniques Used Elsewhere . . . . . . . . . . . 107
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5 Radiation Properties of a Rectangular Microstrip Patch on a Uni-axial Substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1115.2 Equivalent Uniaxial Modeling . . . . . . . . . . . . . . . . . . . . . 113
5.2.1 Hashin-Shtrikman Model . . . . . . . . . . . . . . . . . . . . 1145.2.2 Licktenecker Model . . . . . . . . . . . . . . . . . . . . . . . 114
5.3 Plane Wave Reflection Coefficients for a Grounded Uniaxial Layer . 1195.4 Rectangular Patch Antenna on a Positive Uniaxial Substrate . . . . 1245.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6 Recommendations For Future Work . . . . . . . . . . . . . . . . . . . 1306.1 Arbitrarily Shaped Dielectric Function . . . . . . . . . . . . . . . . 1306.2 Extensions of Effective Medium Theory . . . . . . . . . . . . . . . . 1326.3 Noncommensurate Periodicities . . . . . . . . . . . . . . . . . . . . 1336.4 Extensive Parametric Study of Antenna Performance Near a Periodic
Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
APPENDICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
vii
LIST OF TABLES
Table2.1 Normalized TEx propagation constant kx0a for increasing number of Floquet
modes for f0=fa/c=1, εr=8.9, b=0.3545a, and Nx=40 . . . . . . . . . . . . 352.2 Normalized TEx propagation constant kx0a for increasing number of un-
knowns for f0=fa/c=1, εr=8.9, b=0.3545a, and Np=51 . . . . . . . . . . . 362.3 Normalized TMx propagation constant kx0a for increasing number of Floquet
modes for f0=fa/c=1, εr=8.9, b=0.3545a, and Nx=40 . . . . . . . . . . . . 372.4 Normalized TMx propagation constant kx0a for increasing number of un-
knowns for f0=fa/c=1, εr=8.9, b=0.3545a, and Np=51 . . . . . . . . . . . 384.1 Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relative
permittivity εr=2.56 and dielectric thickness t=0.15λd as a function of fillingfraction b/a for Nx=Ny=Nz=4, Np=Nq=61, φ0=45, θ0=45, and a=0.25λ0 93
4.2 Summary of series transformations . . . . . . . . . . . . . . . . . . . . . . . 954.3 Convergence of higher order Shanks’ transforms of Leibnitz series as a func-
tion of order and number, π=3.14159265358979 . . . . . . . . . . . . . . . . 984.4 Error for various (N, k) Shanks’ transforms for Leibnitz series . . . . . . . . 995.1 Minimum achievable anisotropic ratio (AR) . . . . . . . . . . . . . . . . . . 1185.2 Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relative
permittivity εr=2.56 and dielectric thickness t=0.15λd as a function of fillingfraction b/a for 192 unknowns and 288 unknowns for Np=Nq=61, φ0=45,θ0=45, and a=0.25λ0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.3 Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=2.56 and dielectric thickness t=0.15λd as a function of fillingfraction b/a for 188 unknowns, Np=Nq=61, φ0=60, θ0=30, and a=0.25λ0 123
5.4 Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=6.15 and dielectric thickness t=0.15λd as a function of fillingfraction b/a for 192 unknowns and 288 unknowns for Np=Nq=61, φ0=45,θ0=45, and a=0.25λ0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.5 Dependence of resonant frequency on substrate permittivity . . . . . . . . . 126
viii
LIST OF FIGURES
Figure2.1 One dimensional lattice of dielectric slabs of width b in a periodic lattice
with period a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.2 Full band structure for a normally incident mode in a periodic array of di-
electric slabs with b/a=0.1 and εr=8.9 . . . . . . . . . . . . . . . . . . . . . 192.3 Band structure for one dimensional lattice of dielectric slabs of with filling
fraction b/a=0.5 for (a) εb=εa=13, (b) εb=12, εa=13, and (c) εb=1, εa=13 202.4 Band structure (TMx) for lowest order mode of a periodic array of dielectric
slabs as a function of filling fraction b/a for normal incidence and εr=8.9 . 212.5 Band structure (TMx) for lowest order mode of a periodic array of dielectric
slabs as a function of dielectric constant εr for normal incidence and fillingfraction b/a=0.3545 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Band structure (TMx) for lowest order mode of a periodic array of dielectricslabs as a function of off-axis propagation constant ky0 for filling fractionb/a=0.3545 and εr=8.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 Representative dielectric slab of width b with period a . . . . . . . . . . . . 252.8 Piecewise constant and piecewise linear basis functions used in this work . . 292.9 Band structure (TEx, TMx) of a periodic array of dielectric slabs for b/a=
0.3545, εr=8.9, and off-axis propagation constant ky0=1/a computed usingmethod of moments technique (MoM) with Np=51 and plane wave expansion(PWE) solution with Np=63 . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.10 Microstrip mounted on 1-D periodic dielectric substrate with period a, fillingfraction b/a=0.5, and height t . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.11 Measured frequency response of a microstrip mounted on a 1-D grating and2-D lattice substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.12 Simulated frequency response of HFSS and Libra simulations of a microstripmounted on a 1-D grating . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.13 Microstrip mounted on (a) 1-D grating, (b) finite 1-D grating, (c) 2-D latticeof square holes, and (d) 2-D lattice of circular holes . . . . . . . . . . . . . . 42
2.14 Simulated frequency response of a microstrip mounted on a 1-D grating, finite1-D grating, 2-D lattice of square holes, and 2-D lattice of circular holes . . 43
2.15 Microstrip mounted on two-dimensional periodic dielectric substrate withperiod a, cell-to-cell distance c=a/
√2, hole diameter d, and height t . . . . 43
3.1 Cross-sectional view of two-dimensional array of dielectric rods of diameterb in a periodic square lattice with period a . . . . . . . . . . . . . . . . . . 47
ix
3.2 Unit cell dimensions for (a) rectangular dielectric rods and (b) circular di-electric rods where for the rectangular rod, a and c are the unit cell sizes, band d are the element edge lengths, and for the circular rod, a is the unitcell size and b is the diameter . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3 Unit cell dimensions for (a) rectangular dielectric air columns and (b) circularair columns where for the rectangular column, a and c are the unit cell sizes,b and d are the element edge lengths, and for the circular column, a is theunit cell size and b is the diameter . . . . . . . . . . . . . . . . . . . . . . . 50
3.4 Band structure (TMz) for a normally incident mode in a periodic array ofdielectric rods with b/a=0.3545 and εr=8.9 . . . . . . . . . . . . . . . . . . 54
3.5 Band structure (TEz) for a normally incident mode in a periodic array of aircolumns with b/a=0.6455 and εr=8.9 . . . . . . . . . . . . . . . . . . . . . . 55
3.6 Equivalent structures for Figure 3.2(a) and Figure 3.3(a) . . . . . . . . . . . 553.7 Normalized TMz propagation constant kx0a as a function of d/c for f0=
fa/c=0.5, εr=10.2, ky=0, and b/a=0.2 . . . . . . . . . . . . . . . . . . . . . 563.8 Representative dielectric rod of diameter b in a periodic square lattice with
period a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.9 Band structure (TMz, TEz) of a periodic lattice of dielectric columns with
filling fraction b/a=d/c=0.3545, εr=8.9, Nx=Ny=15, and Np=Nq=31 . . . 663.10 Effective medium theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.11 Exact band structure and effective medium theory approximation for periodic
lattice of dielectric columns with filling fraction b/a=d/c=0.3545, dielectricconstant εr=8.9, and period-to-wavelength ratio α = a/λ0=0.5 for (a) TMz
case and (b) TEz case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703.12 Exact band structure and effective medium theory approximation for peri-
odic lattice of dielectric columns with filling fraction b/a=d/c=0.2, dielectricconstant εr=8.9, and period-to-wavelength ratio α = a/λ0=0.5 for (a) TMz
case and (b) TEz case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 714.1 Doubly periodic dielectric layer with period a in the x direction and period
c in the y direction over a layered medium . . . . . . . . . . . . . . . . . . . 754.2 Coordinate system (kDB) for incident and scattered field . . . . . . . . . . 764.3 Doubly periodic dielectric layered unit cell . . . . . . . . . . . . . . . . . . . 784.4 Plane wave reflection from layered medium . . . . . . . . . . . . . . . . . . 814.5 Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relative
permittivity εr=2.56 and dielectric thickness t=0.15λd as a function of θ0 forNx=Ny=Nz=3, Np=Nq=61, and a=0.1λ0 . . . . . . . . . . . . . . . . . . . 92
4.6 Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=2.56 and dielectric thickness t=0.15λd as a function of θ0 forNx=Ny=Nz=3, Np=Nq=61, and a=0.1λ0 calculated using the multilayeredsolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.7 Relative error as a function of floating point operations for computing variousorder Shanks’ transforms and the direct sum of the inner sum of (4.80) . . 101
4.8 Relative error as a function of floating point operations for computing variousorder Shanks’ transforms, Kummer’s transforms, and the direct sum of theinner sum of (4.80) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
x
5.1 (a) Periodic dielectric layer over a layered medium and (b) equivalent uniax-ial medium with transverse permittivity εt and axial permittivity εz over alayered medium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2 Effective transverse permittivity as a function of filling fraction for εb=10εa 1155.3 Effective transverse permittivity as a function of filling fraction for εa=10εb 1165.4 Uniaxial nature of sapphire modeled by (a) periodic dielectric rods with
ε2=12.1ε1 and b/a=0.95 and (b) periodic dielectric veins with ε2=16.0ε1and b/a=0.30 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.5 Geometry for a rectangular patch on a grounded dielectric substrate . . . . 1255.6 Directivity patterns for reference patch with AR=1 and AR=0.5 . . . . . . 1286.1 Optimized frequency selective layer (volume) . . . . . . . . . . . . . . . . . 1316.2 Dielectric and/or metallic loading . . . . . . . . . . . . . . . . . . . . . . . . 1316.3 Frequency selective supercell . . . . . . . . . . . . . . . . . . . . . . . . . . 1326.4 Higher-order effective medium theory . . . . . . . . . . . . . . . . . . . . . . 1336.5 Noncommensurate periodicities . . . . . . . . . . . . . . . . . . . . . . . . . 133A.1 Loaded and equivalent transmission line circuits . . . . . . . . . . . . . . . . 138B.1 Direct lattice defined using the primitive lattice vectors a and b and the
reciprocal lattice defined using the primitive reciprocal lattice vectors a∗ andb∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
B.2 Irreducible Brillouin Zone (BZ) . . . . . . . . . . . . . . . . . . . . . . . . . 142B.3 Two-dimensional Bravais (or space) lattices . . . . . . . . . . . . . . . . . . 143D.1 Conductor-backed folded slot with periodic dielectric substrate . . . . . . . 154D.2 Simulation unit cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155D.3 Unit cell dimensions for (a) square dielectric rods and (b) circular dielectric
rods where for the square rod, a is the unit cell size, b is the element edgelength, and for the circular rod, a is the unit cell size and b is the diameter 156
D.4 Band-gap plot as a function normalized insert size b/a . . . . . . . . . . . . 156D.5 Full band structure of a square lattice of dielectric columns as a function of
normalized insert size b/a . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157D.6 Irreducible Brillouin Zone (BZ) . . . . . . . . . . . . . . . . . . . . . . . . . 157D.7 Simulated (HFSS) transmission spectra for Γ–X (left) and Γ–M (right) di-
rections and calculated (MoM) band diagram (center) of a two-dimensionalEBG with a=1.2 cm, b=4.8 mm, and εr=10.2 for TM polarization. . . . . . 158
D.8 Top view of conductor-backed folded slot design with periodic dielectric sub-strate (a=1.2 cm, b=4.8 mm, and εr=10.2) where slot dimension in paren-theses refers only to the reference slot design on an homogeneous substrate 159
D.9 Measured and simulated frequency response of conductor-backed slot withperiodic dielectric substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
D.10 Measured normalized E - and H -plane antenna patterns (co- and x-pol) ofreference slot at 9.7 GHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
D.11 Evanescent field structure inside EBG-backed substrate (left) and propagat-ing mode in homogeneously-filled parallel-plate (right) . . . . . . . . . . . . 163
D.12 Measured normalized E - and H -plane antenna patterns (co- and x-pol) ofconductor-backed slot with EBG substrate at 9.7 GHz . . . . . . . . . . . . 164
xi
D.13 Measured normalized E -plane antenna patterns (co-pol) of the reference slotand the slot backed with the periodic dielectric at 9.7 GHz . . . . . . . . . 164
D.14 Axis and pattern cut definitions of the measurement systems . . . . . . . . 166
xii
LIST OF APPENDICES
APPENDIXA Transmission Line Formulation for the One-Dimensional Periodic
Array of Dielectric Slabs . . . . . . . . . . . . . . . . . . . . . . . . . . . 137B Bravais Lattices and the Brillouin Zone . . . . . . . . . . . . . . . . . 140C Impedance Matrix Elements for Two-Dimensional Periodic Struc-
tures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144D Parallel-Plate Mode Reduction In Conductor-Backed Slots Using
Periodic Dielectric Substrates . . . . . . . . . . . . . . . . . . . . . . . 152E Review Of The Method Of Moments . . . . . . . . . . . . . . . . . . . 168
xiii
CHAPTER 1
Introduction
For many years, electromagnetic theorists were provided with two primary means of pre-
dicting electromagnetic phenomena: measurements and analytical solutions. Measurements
of electromagnetic systems provide the scientist/engineer with reams of data and a physical
intuition for the mechanisms involved in producing the response. However, the feasibility
of fabricating new devices for each design change can be time-consuming and, depending
on the architecture or application, expensive. These two limitations are often addressed by
using analytical techniques to predict electromagnetic behavior. Solution techniques includ-
ing asymptotic analysis, variational analysis, integral transforms, contour integration, and
perturbation theory are the staple of analytical electromagnetic prediction and design. Im-
provements in computer speed and memory helped to provide the framework for the rapid
development of computational electromagnetics (CEM) methods including, but not limited
to, the finite element (FE) method, the finite difference methods in both time (FDTD)
and frequency domains (FDFD), the transmission line matrix (TLM) method, and method
of moments (MoM) solutions. CEM techniques have added significantly to the toolboxes
of scientists and engineers alike. In fact, the hybridization of many of the aforementioned
techniques to solve large and complex problems may provide revolutionary new solution
capabilities for the near future.
With the recent advent of general purpose electromagnetics (EM) codes1, a third so-
lution method has been introduced to the technical community, simulation. One might
conclude that the availability of high performance commercial EM software has mitigated
the need to solve specific electromagnetic problems by writing original computer codes.
For example, the late availability of commercial codes that incorporate periodic boundary1An extensive repository of commercial general purpose computational electromagnetic codes and locally
free software is found at the Electromagnetics Library (EMLIB), http://emlib.jpl.nasa.gov.
1
conditions is a prime example of a recent simulation tool that has renewed the interest in
developing new circuit and antenna devices. Certainly, one of the advantages of the general
EM solver is its ability to solve a wide variety of problems. The disadvantage of the general
EM solver is its inability to address specific theoretical or numerical issues that can signifi-
cantly affect the accuracy of a solution. This is not to say that general EM solvers are not
extremely useful for design and have not revolutionized the way that electromagnetics re-
search is being performed. However, every CEM code writer will attest to the many nuances
and small singular changes that must be included in an original numerical code to achieve
maximum performance. Complex, real-world problems are not always amenable to gen-
eral purpose solutions and often require customized solutions and programs to solve them.
Computational electromagnetics which includes electromagnetic theory, sophisticated ana-
lytical techniques, and numerical methods is used to characterize and analyze the merits
and demerits of a new design.
The rapidly growing field of wireless communications is providing new opportunities to
develop novel structures that will enhance or even replace existing circuits and antennas.
Devices that incorporate periodicity as a key feature of the design are promising unri-
valed performance in microwave circuits and antennas. Periodic electromagnetic structures
are commonplace in many of the items we rely on every day – from the magnetron in a
microwave oven to the ultraviolet radiation protection provided by polarized sunglasses.
Commenting on the recent inundation of applications for periodic structures, Maddox [57]
remarks, “If only it were possible to make materials [photonic band-gap materials] in which
electromagnetic waves cannot propagate at certain frequencies, all kinds of almost-magical
things would happen.” It is important, nevertheless, to dispel some of the myths about
what photonic band-gap (PBG) materials are and are not. PBGs are periodic dielectric
and/or metallic structures that when designed and implemented correctly can improve the
performance of specific devices. PBGs are not magical structures that defy basic laws of
physics and have only of late appeared in the literature.
Photonic band-gap materials, sometimes referred to as electromagnetic band-gap (EBG)
materials, electromagnetic crystals (EC) or photonic crystals, are inhomogeneous structures
composed of periodic regions of material with a specific permittivity immersed in a homoge-
neous background of different permittivity. These artificial, composite structures have been
found to have unique properties that are advantageous in applications involving semicon-
ductor integrated circuits. Researchers from the diverse fields of classical electromagnetics,
solid-state physics, optics, material science, condensed matter physics, and semiconductor
2
physics, are actively contributing to the base of electromagnetic crystal knowledge. Active
areas of electromagnetic and photonic crystal research include but are not limited to mi-
crowave and millimeter-wave antenna structures, quasi-optical microwave arrays, photonic
crystal integrated circuits, high- and low-Q electromagnetic resonators, quantum optical
electromagnetic cavity effects, and optical nano-cavities. In addition, sonic band-gap ma-
terials or artificial acoustic crystal substrates, are being developed and could impact sonar.
From solid-state theory, we know that semiconductors allow electron conduction with-
out scattering only for electrons that have energies within a specific range of energy, often
termed band-gaps. Electromagnetic wave propagation in periodic dielectric media is analo-
gous to electron-wave propagation in semiconductor crystals. Although fundamentally dif-
ferent propagation mechanisms are involved, preliminary results suggest that at microwave
and millimeter-wave frequencies the propagation characteristics of these crystals can be ma-
nipulated by carefully designing and fabricating structures composed of regions of differing
dielectric constants. Early work by Yablonovitch [125] successfully demonstrated that light
propagation could be inhibited in certain frequency gaps in special photonic band-gap crys-
tals (PBGs). Scalar-wave-based theories were developed to determine a suitable candidate
for the first three-dimensional (3-D) photonic crystal. However, it became apparent that
scalar-wave-based solutions were inadequate for the task. Vector-wave-based solutions im-
plemented using the plane wave expansion (PWE) method for Maxwell’s equations, being
developed concurrently by Leung et al. [52], Zhang et al. [138], and Ho et al. [36], offered
the potential of finding a true photonic band-gap structure.2 Indeed, soon thereafter, a
structure that possesses a full photonic band-gap was identified. Theoretical work by Ho et
al. [36] confirmed the existence and structure of the first (3-D) photonic crystal, the di-
amond dielectric structure. Concurrently, band structures for two-dimensionally periodic
structures were being calculated in a similar manner by Plihal et al. [73]. However, it
was McCall et al. [60] who were the first to calculate and measure microwave propagation
and reflection in a two-dimensional (2-D) array of low-loss high-dielectric-constant cylin-
ders. With the confirmation of realizable 2-D and 3-D photonic crystal structures, attention
shifted to applying these materials in new designs and architectures.
Reviews of early photonic band-gap research can be found in special issues of the Journal
of the Optical Society of America B [1] and the Journal of Modern Optics [2]. Recently, a
book outlining photonic band theory and covering a wide range of photonic crystal applica-
tions was published by Joannopoulos [38]. Although previous research had focused almost2Earlier experimental work had only verified a pseudo-band-gap, not a true photonic band-gap. See [126].
3
exclusively on optical and quasi-optical applications for the PBG materials, it quickly be-
came apparent that numerous applications for similar periodic structures in the microwave
and infrared frequency range were being developed. Consequently, a joint special issue
of the IEEE Transactions on Microwave Theory and Techniques [3] and the Journal of
Lightwave Technology [4] was published to catalog the design, synthesis, and application of
electromagnetic crystal structures begin developed in the microwave and millimeter-wave
community. Concurrently, a special issue that focused on the particular theoretical and
numerical aspects of photonic band-gap structure research was released by the journal Elec-
tromagnetics [5].
It is the purpose of this work to analyze periodic structures to determine effective and re-
alizable uses for microwave circuits and antennas. In Chapter 2, full-wave integral equation
(IE)/method of moments (MoM) solutions are developed and implemented to determine the
band structure (propagating modes) of a periodic, one-dimensional inhomogeneous dielec-
tric region. A single electric field integral equation (EFIE) or coupled EFIEs incorporating
both the periodic free-space Green’s function (1-D) and equivalent electric polarization cur-
rents are formulated to determined the quantities of interest. Subsequently, the geometry
and integral equation(s) are discretized, and the method of moments is used to numerically
solve the resulting matrix equation. A nontrivial solution for the fields requires the matrix
determinant to be zero, which results in a characteristic equation. The eigenvalues (prop-
agation constants) are obtained from the roots of this equation. For a lossless structure,
the propagation constants of a guided wave are real numbers. However, in the stopbands,
the propagation constants are complex-valued. To validate the solution obtained through
the use of the IE/MoM approach, the solution of the exact eigenvalue equation for the
one-dimensional periodic problem is also obtained using the plane wave expansion (PWE)
method.
In Chapter 3, similar approaches as implemented in Chapter 2 are used to extend the
solution to two-dimensionally periodic media. Additional integral equation (IE)/method of
moments (MoM) codes are developed and implemented, but this time to determine the band
structure (propagating modes) of a periodic, two-dimensional inhomogeneous dielectric re-
gion. A single EFIE or coupled EFIEs are formulated using a periodic Green’s function
(2-D) along with equivalent electric polarization currents, and the method of moments is
again used to numerically solve the resulting matrix equation. The exact eigenvalue equa-
tion for the two-dimensional periodic problem is solved as before using the PWE method
to verify the IE/MoM solution. To simplify the computation of the band structure for two-
4
dimensionally periodic media, effective medium theory is applied in Section 3.4 to reduce
the two-dimensional periodic structure to a one-dimensional equivalent structure. This use-
ful technique can be applied when the period of the structure is much smaller than the
wavelength.
Recently, Ansoft, a leading developer of computational electromagnetic software, ex-
tended the capabilities of their 3-D commercial finite element software package to include
linked boundary conditions, their particular implementation of periodic boundary condi-
tions. The new capability enables Ansoft to perform new calculations on infinite arrays
similar to the waveguide simulator concept developed in the 1960s to determine array per-
formance without building an entire array assembly. A number of modeling methodologies
including direct-transmission methods, dispersion diagram methods, and reflection phase
analysis methods were examined to test the new linked boundary conditions [77]. To val-
idate the accuracy of the direct-transmission method, Ansoft chose to repeat the solution
developed in Chapter 3 and published by this author et al. in [91] to determine the band
structure (propagating modes) of a periodic, two-dimensional inhomogeneous dielectric re-
gion. The confidence in the accuracy of the full-wave integral equation (IE)/method of
moments (MoM) solution developed in this work provides a standard with which the new
linked boundary condition solution could be compared.
A number of researchers have designed electromagnetic crystal structures for use in
planar antenna and circuit applications, particularly for use as reflectors in planar dipole
antenna structures [18, 21, 17, 41, 95, 97]. Planar antennas are ideal for use in many
wireless networks including personal communications systems (PCS) and mobile satellite
communications. Microstrip antennas are of interest to many wireless users because of
their low aerodynamic profile (survivability/durability issue), light weight (volume/size),
and low cost. Microstrip antennas also have the advantage of being both conformal and
easily integrated into thin-film circuits. Unfortunately, microstrip antennas have relatively
low gain and narrow bandwidths due to substrate surface wave formation. By designing
special artificial substrates, surface wave formation may be reduced or even eliminated in
some planar antennas. This would dramatically increase both the available bandwidth and
gain to levels usually reserved for non-planar antennas. To this end, periodic structures
may serve as the material that can change the physical properties of substrates used in
fabricating planar circuits and antennas.
Only recently, have researchers developed periodic structures for application in slotted
antennas. Leung et al. [53] measured the radiation patterns of a slot antenna placed on a
5
layer-by-layer photonic band gap crystal. For planar antennas operating at a frequency in
the band-gap of the three-dimensional PBG crystal, energy which would have been radiated
into the substrate is reflected. However, at the interface between the PBG and the air,
the period of the PBG is broken and a parasitic mode (surface state) can exist. These
surface states decrease the efficiency by stripping power away from the radiating element.
However, by fabricating a resonant slot over a reflecting back plate and filling the resulting
parallel-plate with an appropriately designed artificial electromagnetic band-gap structure,
noticeable enhancements in both radiation pattern and bandwidth can be achieved using
a significantly lower profile than traditional designs. This design is detailed extensively in
Appendix D of this work.
As the need for very wideband, omni-directional antennas for use in mobile communica-
tion networks grows at an increasing rate, so does the need to develop custom solutions (and
the resulting codes) for novel antenna architectures and designs. The salient properties of
the antenna, such as its radiation pattern and efficiency, its input impedance, bandwidth
and gain, are all determined by the antenna’s electrical size, physical configuration, and the
environment in which it is located. In order to optimize the performance of the antenna, a
number of design parameters must be considered. Element shape, resonant frequency, gain,
and bandwidth of the individual element must be properly chosen. For planar circuits and
antennas, substrate characteristics must be designed that take into account the effect of
dispersion on the effective dielectric constant and the effect of substrate surface wave for-
mation. Other concerns, including the environmental effects of temperature, humidity, and
aging; the mechanical concerns of vibration effects and durability; and the ease of conforma-
bility/machinability must also be addressed. Additionally, feed structures must be designed
to minimize insertion loss, to maximize space utilization, and to minimize discontinuities.
Until recently [77], general EM solvers were unable to solve even simple periodic struc-
tures such as artificial dielectrics or large antenna arrays. Consequently, custom codes are
being developed to characterize important parameters such as the reflection coefficient, ra-
diation pattern, and radiation efficiency for planar antennas mounted on layered periodic
substrates. Significant attention has been directed to developing an entire class of structures
termed frequency selective surfaces [66]. A frequency selective surface is usually designed to
transmit or reflect electromagnetic radiation within a band of frequencies (or within specific
angular bands) from propagating through the surface. This is often accomplished by placing
metallic patch elements or aperture elements, such as annular rings, crossed slots, crossed
dipoles, or jerusalem crosses, in a periodic fashion on the surface of a dielectric layer. Ra-
6
diation near the resonant frequency of the element will be transmitted or reflected through
the surface as the design permits. Applications for such structures include broadband mi-
crowave antenna radome design, narrow-band frequency selective surfaces, and polarizers.
These structure are often evaluated using a generalized scattering matrix technique [135]
that cascades the propagation through the different layers. Less attention has been fo-
cused on three-dimensional frequency selective layers (volumes). The solutions developed
in Chapters 4 and 5 of this work are amenable to FSS design and have applications both
for substrate and superstrate design.
Recent numerical work in determining the microwave scattering from periodic material
implants in a layered medium has been done by Sarabandi [85], Sarabandi et al. [87],
Tsay et al. [114], Yang [130, 134], and Yang et al. [135, 136]. In Chapter 4, the solution
by Sarabandi [84] for the two-dimensional scattering from an inhomogeneous periodic layer
above a half-space layered medium is extended to general three-dimensional scattering from
an inhomogeneous doubly periodic layer above a half-space layered medium. The solution
of this problem is derived and implemented using a full-wave integral equation/method of
moments approach similar to the formulation for the one- and two-dimensional periodic
structures of Chapters 2 and 3.
Of particular concern in the solution of periodic structures is the convergence of the
resulting Floquet series. Even though the contribution of the off-plane periodic elements
converges very quickly, the convergence of the on-plane periodic elements is notoriously slow.
In order to compute the impedance matrix elements in a reasonable amount of time, various
series acceleration techniques and transformations have been suggested [42]. Summation
acceleration techniques convert a slowly converging series to a rapidly converging one by
allowing the series to be transformed into a second series that converges to the same limit but
does so in a rapid fashion. For the problems of interest in Chapter 4, the combination of a
Poisson transformation [70] and a Shanks’ transformation [90] are successfully implemented
to improve both the speed and the accuracy of the impedance matrix element computations.
This is the first known successful implementation of a Shanks’ transformation applied to
each of the series in the double sum found in the planar two-dimensional, periodic free-
space Green’s function. Additionally, Kummer’s method [51] is applied to specific series to
compare the convergence rate of these different series acceleration techniques. A complete
treatment of these and other series acceleration techniques is found in Section 4.5.
Of interest to the applied microwave community is the implementation of a microstrip
patch antenna on a doubly periodic dielectric layered medium. The solution procedure
7
developed in Chapter 4 to determine the scattering from doubly periodic dielectric layered
medium is used to validate an equivalent model that replaces the periodic layer with an
equivalent uniaxial layered medium. The ability of the model to emulate the periodic layer is
validated using plane wave reflection coefficients for various combinations of filling fractions,
angles, and permittivities and the limitations of the model are addressed. Subsequently, the
solution for a rectangular microstrip patch element radiating over the equivalent uniaxial
layered medium is carried out accurately and efficiently using a commercial finite element
package packages that incorporates anisotropic substrates. Conclusions about the effect of
the anisotropy on various antenna parameters such as resonant length, pattern shape, and
array coupling are drawn and numerical examples that illustrate the salient features of the
single patch antenna integrated on a uniaxial substrate are presented in Chapter 5.
Traditionally, microstrip patch antennas have been integrated on relatively low per-
mittivity substrates in order to improve antenna performance. Integrating the antenna
on higher permittivity substrates is preferred to minimize circuit size and spurious radia-
tion [75, 89] but at the cost of confining the potential radiating energy even more tightly.
This trade-off between good antenna performance and good circuit performance is a key
design feature found in many microstrip antenna designs. Although a good deal of atten-
tion has focused on integrating microstrip patches on homogeneous substrates, many of the
practical substrates with higher permittivities in use today such as sapphire have a signifi-
cant amount of (uniaxial) anisotropy. The primary effect of anisotropy on rectangular patch
antenna design is the change in its resonant length (frequency). This is significant because
of the narrow bandwidth of the patch itself. The relatively large shift in resonant frequency
produced in many of the modern substrates may actually force a rectangular patch de-
signed to operate at a specific frequency to radiate outside of the antenna bandwidth [74].
Additionally, anisotropic effects are found that shape the radiation pattern of the patch
and thus in an array configuration, the coupling to other elements. Because uniaxial sub-
strates are often expensive to manufacture and have limited flexibility for design, uniaxial
substrates can be emulated (and easily fabricated) by incorporating periodic inclusions in
an otherwise homogeneous substrate. The doubly periodic structure can be constructed us-
ing simple milling or etching techniques from simple inexpensive, homogeneous substrates.
This is significant because some common uniaxial materials such as sapphire that are ex-
pensive to grow [6] can be “artificially” replicated easily and inexpensively. Additionally,
the artificial nature of the periodic uniaxial substrate permits the creative design of new
substrates with the expanded freedom of anisotropic ratio, background permittivity, and/or
8
fabrication technique.
A number of potential applications for the solution techniques developed in this work
are outlined in Chapter 6. The specification of an arbitrarily shaped dielectric is one inter-
esting design feature that can be optimized in the solution of the one- and two-dimensional
periodic media. Additionally, lattices of dielectric elements with differing dielectric con-
stants can be included to provide additional freedom for the design. Two potential appli-
cations for effective medium theory that promise to be of value are the extension of EMT
to off-axis propagation in two-dimensional lattices and to out-of-plane propagation in two-
dimensional lattices. A number of new applications can be developed using the solution
technique implemented in Chapters 4 and 5 including new frequency selective surfaces (vol-
umes), incorporating separate periodicities (non-commensurate periodicities) for each layer,
incorporating material implants within each layer of differing relative permittivity (dielec-
tric and/or metallic loading), and the extension of the solution to large planar antenna
elements. Perhaps the most significant development would be incorporation of many of the
above suggestions to eliminate surface wave formation in planar antenna applications.
9
CHAPTER 2
One-Dimensional Periodic Dielectric Structures
2.1 Introduction
One-dimensional periodic electromagnetic structures are commonplace in many of the
items we rely on every day – from the magnetron in a microwave oven to the ultraviolet
(UV) radiation protection provided by polarized sunglasses. In particular, the field of optics
is replete with one-dimensional periodic structures. How periodicity is incorporated into
each device is what distinguishes one design from another. For example, the direction of
period in some structures is in the direction of stratification, whereas in others, the period
is in a transverse plane to the direction of stratification. Bragg mirrors (reflectors) are an
excellent example of periodicity in the direction of stratification. The highly reflective coat-
ings used in Bragg reflectors are produced by “sandwiching” layers of differing dielectric
material with highly reflective properties. A recent example of active research in this area
has been reported in [22] where total omni-directional reflections from one-dimensional
dielectric lattices have been reported. Other application in this class include specialized
thin-film waveguiding, quarter-wave stacks, and folded Solc filters [137]. Alternating layers
of metal and dielectric are being developed constantly that provide almost unbelievable
new applications. Transparent metallic structures have been developed that permit the
transmission of light over a tunable range of frequencies, effectively blocking both UV and
infrared and lower frequencies from propagating through [88]. Optical and quasi-optical
researchers have produced new sensor and eye protection devices, heat reflecting windows,
better ultraviolet blocking films, transparent electrodes for light emitting diodes, and light
crystal displays. The second class of periodicity, corrugation along the surface of a dielectric
medium, is used in devices such as diffraction gratings and high-reflectance distributed-
10
Bragg-reflector (DBR) lasers. Diffraction gratings are the most commonly known appli-
cation of corrugations and are used to provide certain spectral or angular characteristics
for reflection and transmission of electromagnetic radiation. Other interesting applications
include distributed-feedback (DFB) lasers where an active layer provides gain, TM-to-TE
mode conversion, and forward-backward mode conversion.
For microwave and RF engineers, the gridded traveling wave tube (GTWT) amplifier is
a prime example of harnessing the potential power of a periodic device. The GTWT uses
a helical waveguide to convert the high-energy of an electron beam to microwave energy
and is implemented on almost every airborne radar in the world. At lower frequencies, the
remote sensing community is actively studying naturally occurring one-dimensional periodic
structures such as traveling ocean waves and periodic geological structures and man-made
periodic structures such as periodic vegetation [112].
A fundamental understanding of how electromagnetic fields behave in a periodic medium
is required not only to apply these unique properties correctly but also to critique designs
or theories where they are applied incorrectly. The foundation for electromagnetic wave
propagation in periodic media is provided in Section 2.2. The phase constants for elec-
tromagnetic waves propagating in a one-dimensional lattice of dielectric slabs in an air
background is found explicitly using two different techniques. The first solution, detailed
in Section 2.3, incorporates the use of a Fourier series representation for the periodic field
and is determined by solving the one-dimensional wave (differential) equation for periodic
media. The second solution is found by deriving an integral equation for the periodic field
and numerically solving the resulting linear system and is presented in Section 2.4. A
number of representative structures are used in the chapter to show the behavior of elec-
tromagnetic fields in one-dimensional periodic media and observations about the salient
features of each are discussed. Although theoretical band structures are useful, actual re-
alizable devices that incorporate a design and that can be fabricated also provide valuable
insight for understanding the propagation mechanisms at work. An example solution for
the microstrip excitation of a one-dimensional periodic dielectric structure is presented in
Section 2.5. Since the fields produced by the quasi-TEM mode of the microstrip (for high
dielectric constant) are confined relatively close to the microstrip itself, the dielectric sub-
strate is modelled as a hi-Z, low-Z filter. The filter design is shown to match closely to
that of periodic plane wave propagation in a lattice of dielectric slabs and to the measured
response of the microstrip line itself.
11
2.2 Fundamental Concepts
2.2.1 Maxwell’s Equations
The fundamental equations that form the foundation for electromagnetic theory are
Maxwell’s equations. Written in differential form, these equations are
∇× E = −∂B∂t
(2.1a)
∇× H = J+∂D∂t
(2.1b)
∇ · D = ρ (2.1c)
∇ · B = 0 (2.1d)
where E is the electric field intensity, B is the magnetic flux density, H is the magnetic
field intensity, and D is the electric flux density. The electric current density J and electric
charge density ρ are the sources of the electromagnetic fields.
For linear and isotropic media, E and D and B and H are related by the constitutive
relations
D = εrε0E (2.2a)
B = µrµ0H, (2.2b)
where ε0, µ0, εr, and µr are the free-space permittivity, free-space permeability, relative
permittivity, and relative permeability, respectively.
The equation of continuity provides the relation between the electric charge density and
electric current density through
∇ · J = −∂ρ∂t
(2.3)
which is statement of the conservation of charge.
The remaining fundamental equation, Lorentz’s force equation, determines the total
electromagnetic force on a charge q to be
F = q (E + u× B) (2.4)
where u is the velocity of the moving charge.
The eight equations, comprised of the four Maxwell’s equations (2.1), together with the
constitutive relations (2.2), the equation of continuity (2.3), and Lorentz’s force equation
(2.4), provide the necessary framework to predict all macroscopic electromagnetic interac-
tions. Although the four equations (2.1a)–(2.1d) are not independent, equations (2.1c) and
(2.1d) can be derived from (2.1b) and (2.1a) using (2.3).
12
2.2.2 Propagation of Waves in a Periodic Medium
One of the many interesting problems in mathematical physics is the solution of the wave
equation in periodic media. The solution of this type of equation is usually derived using
a form of Floquet’s theorem. Floquet’s theorem, as it was originally presented [120], deter-
mined a particular periodic solution of Mathieu’s equation, the equation of wave motion.
Subsequently, this solution was extended for Hill’s equation, a generalization of Mathieu’s
equation, for any periodic function, providing the basis for one of the great successes in
applied physics: quantum band theory. Band theory, which describes the properties of
electrons in a periodic potential due to the atomic arrangement of atoms in a crystal, is
the foundation for the understanding of electronic transport in metals, semiconductors, and
insulators. The solution of the periodic potential problem can be expressed in mathematical
form as Bloch’s theorem. Bloch’s theorem states that the eigenfunctions of the Schrodinger
equation for a periodic potential are the product of a plane wave and a function which has
the same period as the periodic potential [98]. A number of similarities can be seen in the
solution of electron wave propagation in semiconductors and electromagnetic wave propa-
gation in periodic dielectric media, and Bloch’s theorem can be extended to electromagnetic
wave propagation in periodic media.
For propagating modes in a periodic dielectric structure, the field in an adjacent cell is
related by a complex constant. That is,
E(x, y, z + a) = E(x, y, z) e−jβ0a (2.5)
where E(x, y, z) is a periodic function of z with period a and β0 is the phase constant in
the z direction. If the propagating mode in the structure has the form
E(x, y, z) = Ep(x, y, z) e−jβ0z (2.6)
where Ep(x, y, z) is a periodic function of z with period a, then
E(x, y, z + a) = Ep(x, y, z + a) e−jβ0(z+a). (2.7)
Since Ep(x, y, z) is periodic with period a,
Ep(x, y, z + a) = Ep(x, y, z). (2.8)
Substituting (2.8) into (2.7) yields
E(x, y, z + a) = Ep(x, y, z) e−jβ0(z+a). (2.9)
13
But, using (2.6),
E(x, y, z + a) = E(x, y, z) e−jβ0a, (2.10)
which is exactly (2.5), a statement of Floquet’s (or Bloch’s) theorem.
From the theory of Fourier series, the periodic electric field with period a can be ex-
panded as a periodic function in x and prescribed phase constant β0 and is given by
E(x, y, z) =∑n
En(x, y) e−j2πn
az e−jβ0z =
∑n
En(x, y) e−jβnz (2.11)
where
En(x, y) =1a
a∫0
En(x, y, z) ej2πn
az dz
are the coefficients that serve to represent the dependence on x and y, and
βn = β0 +2πna
is the phase constant of the nth harmonic.
Thus, the field in a periodic structure can be expanded in an infinite set of harmonics
through Floquet’s theorem, each with frequency f and propagation constant βn. The phase
velocity νn of the nth harmonic is
νn =ω
βn=
ω
β0 + 2πna
. (2.12)
For slow-wave structures, the phase velocity of the nth harmonic can be less than the free-
space velocity, or νn < c. However, the group velocity, νg, found from
νg =1
∂βn/∂ω=
1∂(β0 + 2πn
a
)/∂ω
=∂ω
∂β0(2.13)
is independent of n.
2.3 Plane Wave Expansion Method
2.3.1 Analytical Techniques
A straight forward solution to the exact eigenvalue equation for the one-dimensional
periodic problem is obtained through the use of the Fourier series. A representative one-
dimensional periodic array of dielectric slabs with period a and dielectric insert width b is
14
x
y
ba
E(H)
φz
Figure 2.1: One dimensional lattice of dielectric slabs of width b in a periodic lattice withperiod a
illustrated in cross section in Figure 2.1. If the electric field does not have a component
in the x -direction (the direction of period or stratification), the mode is denoted TEx or
horizontal. If the magnetic field does not have a component in the x -direction, the mode is
denoted TMx or vertical. For normal incidence, the modes are degenerate (TEx is equivalent
to TMx).
Transverse Electric (TEx) Case
The electric field can be expanded as a periodic function of plane waves in x with period
a and prescribed propagation constant of kx0 and is given by
E(x, y) = zEz(x, y) = zEp(x) e−jkx0x e−jkyy (2.14)
where Ep(x) is the periodic electric field that propagates only in the xy-plane, i.e. kz = 0,
without loss of generality. Since the electric field must satisfy the wave equation, we now
apply the operator(∇2
xy + k2)to Ez(x, y) of (2.14) noting that the dielectric constant is a
function of x
∇2xyEz(x, y) + k
20εr(x)Ez(x, y) = 0. (2.15)
15
Assuming the parallel slabs are infinite in the y and z directions, (2.15) can be simplified
to
− d2
dx2Ez(x, y) + k2
yEz(x, y) = k20εr(x)Ez(x, y). (2.16)
The periodic electric field is expanded in a Fourier series in x with unknown coefficients an
which serve to represent the dependence on y
Ep(x) =∑n
an e−j 2πn
ax. (2.17)
Since the dielectric function is also periodic, it is appropriate to expand it in another Fourier
series with coefficients bm
εr(x) =∑m
bm e−j 2πm
ax. (2.18)
Substituting the Fourier expansions for the field and the dielectric into (2.16) and carrying
out the algebraic operations, we obtain
∑n
[(2πna
+ kx0
)2
+ k2y
]an e
−j 2πnax = k2
0
∑n
∑m
anbm e−j 2πm
ax e−j
2πnax. (2.19)
In order to determine the unknown coefficients an and bm, (2.19) is multiplied by an orthog-
onal function and integrated over one unit cell which produces a Kronecker delta function
for a specific index
∑n
[(2πna
+ kx0
)2
+ k2y
]anδ
(2πpa
− 2πna
)= k2
0
∑n
∑m
anbmδ
(2πpa
− 2πma
− 2πna
).
(2.20)
The convolution in equation (2.20) is easily cast into the following general matrix form[(2πna
+ kx0
)2
+ k2y
]an = k2
0
∑m
bn−mam (2.21)
where
bn−m =1a
b/2∫−b/2
(εr − 1) e−j2π(n−m)
ax dx+
1a
a/2∫−a/2
(1) e−j2π(n−m)
ax dx
=b
a(εr − 1) sinc
π(n−m)ba
+ δn−m. (2.22)
A generalized linear eigensystem problem is represented by Ax = λBx where A and B
are n×n matrices. The value λ is an eigenvalue and x = 0 is the corresponding eigenvector.
The propagating modes in the TEx case are solutions of the generalized linear eigensystem
in (2.21).
16
Transverse Magnetic (TMx) Case
The solution for the TMx case is similar to the TEx case with the significant difference
that the electric field in (2.14) is replaced by the magnetic field in the wave equation
∇×
1εr(x)
∇× zHz(x, y)+ k2
0zHz(x, y) = 0. (2.23)
In the derivation of (2.15), the curl operator is applied to both sides of (2.1a). The expression
for the magnetic field found from (2.1b) is then substituted into the resulting equation which
when simplified, yields (2.16). However, when the curl operator is applied to (2.1b) first,
as it is in the TMx case, the operator acts on both the field and the periodic dielectric
function. Careful attention must be directed to correctly evaluating the specified operators.
The resulting equation that must be solved is
1εr(x)
z · ∇ ×∇× zHz(x, y) + z · ∇
1εr(x)
×∇× zHz(x, y) = −k2
0Hz(x, y). (2.24)
The Fourier series representation of the periodic magnetic field with unknown coefficients
an is
Hp(x) =∑n
an e−j 2πn
ax (2.25)
and for the inverse of the dielectric function, it is
1εr(x)
=∑m
bm e−j 2πm
ax (2.26)
with coefficients bm different from coefficients bm in (2.18) for the TEx case. By substituting
the expressions for the field and dielectric function into (2.24), carrying out the curl and
gradient operations, and simplifying the resulting expression yields
∑m
bme−j 2πm
ax
− d2
dx2
∑n
an e−j 2πn
ax e−jkx0x + k2
y
∑n
an e−j 2πn
ax e−jkx0x
− d
dx
∑m
bme−j 2πm
ax d
dx
∑n
an e−j 2πn
ax e−jkx0x = −k2
0
∑n
an e−j 2πn
ax e−jkx0x. (2.27)
If the derivatives in (2.27) are evaluated and integrated over one unit cell, then the resulting
expression can be cast into a different matrix equation to solve for the unknown eigenvalues
∑m
bn−mam
[(2πna
+ kx0
)2
+ k2y −
2π(n−m)a
(2πna
+ kx0
)]= −k2
0an (2.28)
17
where
bn−m =1a
b/2∫−b/2
(1εr
− 1)e−j
2π(n−m)a
x dx+1a
a/2∫−a/2
(1) e−j2π(n−m)
ax dx
=b
a
(1εr
− 1)sinc
π(n−m)ba
+ δn−m. (2.29)
An ordinary linear eigensystem problem is represented by the equation Ax = λx where
A denotes an n × n matrix. The propagating modes in the TMx case are solutions of the
ordinary eigensystem problem in (2.28).
2.3.2 Matrix Solution of the Eigensystem
The solution of (2.21) or (2.28) is amenable to fast computation using the commercial
software package MATLAB. The resulting eigenvalues of the matrix are the squares of the
frequencies of the propagating modes in the structure. The solution of the frequencies of
the propagating modes in the structure are found for specific values of phase shift kx0a ∈[0, 2π]. Of course, if an infinite number of Floquet modes are used in the solution, then an
infinite number of frequencies (space harmonics) will satisfy (2.21) or (2.28). However, the
limitation of a finite computer memory requires that only a finite number of frequencies
can be found for a given phase shift. If the resulting eigenvalues are sorted from largest
to smallest, the last eigenvalue yields the lowest space harmonic that will propagate for a
given phase shift. Additional propagating modes are found from the ascending eigenvalues.
In Figure 2.2 is plotted a finite portion of the Brillouin zone (BZ) diagram1 for a normally
incident TEx mode (ky=0) in a periodic array of dielectric slabs with filling fraction b/a=0.1
and relative dielectric constant εr=8.9. The filling fraction is defined as the volumetric
ratio of dielectric material to the total volume. The band structure is shown for normalized
propagation values kx0a ∈ [−3π, 3π]. The dark regions are the stop bands. Seeing that
the diagram repeats itself every 2π, only normalized propagation values between 0 and π
need to be calculated to determine all of the propagating modes in the structure. This
region is often termed the irreducible Brillouin zone [38, 44]. For one-dimensional periodic
structures, this a trivial determination. For the two-dimensional structures encountered in
Chapter 3, the irreducible Brillouin zone becomes more complicated. A complete treatment
of the Brillouin zone for two-dimensional lattices is given in Appendix B.1For band structures in the literature, the normalized propagation constant is plotted on the abscissa
and the normalized frequency is usually plotted on the ordinate.
18
Normalized propagation constant (k xa)
-9.42 -6.28 -3.14 0.00 3.14 6.28 9.42
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.5
1.0
1.5
2.0
2.5
Figure 2.2: Full band structure for a normally incident mode in a periodic array of dielectricslabs with b/a=0.1 and εr=8.9
The width of the stopbands and passbands is a function of the relative dielectric con-
stant εr and the filling fraction b/a of the one-dimensional lattice. To illustrate this fact,
representative band structures for a one dimensional lattice of dielectric slabs with filling
fraction b/a=0.5 for various combinations of dielectric contrast are shown in Figure 2.3
for kx0a ∈ [−π, π]. When the permittivity difference between the two regions is negligible
(Figure 2.3(a)), the wave propagates uninhibited for all frequencies, i.e., no stopgaps. If a
small dielectric contrast is introduced between the two materials (Figure 2.3(b)), small gaps
can form. However, the required interference effect between the periods is simply too small
to produce a gap of any consequence. Large contrasts in dielectric constant can produce
significant gaps as seen in Figure 2.3(c).
A number of simple checks can be performed to determine whether the solution for
the periodic slabs is correct in addition to checking the calculated values against published
data. For convenience, the slab is placed in the center of the unit cell and the propagation
constant is determined for a specific combination of electrical and geometrical parameters.
However, the particular location of the slab within the unit cell should have no effect on
the propagating structure. In particular, this is found to be true for two small strips of
dielectric material each half the original dielectric insert width located at the edges of the
unit cell. Other observations are drawn from the calculations by reducing the geometrical
and electrical parameters to special cases, including the following: (i.) as the filling frac-
19
Normalized propagation constant (k xa)
-3.14 -1.57 0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.05
0.10
0.15
0.20
0.25
0.30εb=13,εa=13
(a)
Normalized propagation constant (k xa)
-3.14 -1.57 0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.05
0.10
0.15
0.20
0.25
0.30εb=12,εa=13
(b)
Normalized propagation constant (k xa)
-3.14 -1.57 0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
εb=1,εa=13
(c)
Figure 2.3: Band structure for one dimensional lattice of dielectric slabs of with fillingfraction b/a=0.5 for (a) εb=εa=13, (b) εb=12, εa=13, and (c) εb=1, εa=13
tion decreases to zero, the calculated propagation constant kx0 approaches the minimum
value, the free-space propagation constant k0 (Figure 2.4); (ii.) as the filling fraction in-
creases to 1, the calculated propagation constant approaches the maximum value, k0√εr
(Figure 2.4); and (iii.) as the relative dielectric constant εr is reduced to 1, the calculated
propagation constant approaches the minimum value, the free-space propagation constant
k0 (Figure 2.5).
In Figure 2.6, the band structure (TMx) for the lowest order mode of a periodic array
of dielectric slabs with filling fraction b/a=0.3545 and εr=8.9 is shown for different values
of off-axis propagation. The values for the filling fraction and the relative dielectric con-
stant are obtained from a two-dimensional structure in [64, 129]. In order to determine the
correctness of the results obtained for the two-dimensional periodic structure, the filling
fraction in one lattice direction is allowed to increase until an equivalent one-dimensional
20
Normalized propagation constant (k xa)
0.00 0.79 1.57 2.36 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.1
0.2
0.3
0.4
0.5b/a=0.05b/a=0.1b/a=0.25b/a=0.5b/a=0.75k0
k0εr1/2
Figure 2.4: Band structure (TMx) for lowest order mode of a periodic array of dielectricslabs as a function of filling fraction b/a for normal incidence and εr=8.9
Normalized propagation constant (k xa)
0.00 0.79 1.57 2.36 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.1
0.2
0.3
0.4
0.5er=2
εr=4
εr=9
εr=16
εr=25
k0
Figure 2.5: Band structure (TMx) for lowest order mode of a periodic array of dielectricslabs as a function of dielectric constant εr for normal incidence and fillingfraction b/a=0.3545
periodic structure is obtained. Thus, the solution of the one-dimensional equivalent struc-
ture is needed. Similar results for the band structure of the lowest order TEx mode are
not shown. Note that for normal incidence (ky0 = 0), the first (and lowest) band does not
open up until the normalized frequency f0 ≥ 0.2. However, when the mode is allowed to
propagate with an off-axis component, i.e. φ0 = 0), the lowest band opens up for small
21
normalized frequencies. As the off-axis propagation constant increases, the bandwidth of
the lowest stopband increases and the bandwidth of the lowest passband decreases. In the
limit as the off-axis propagation increase, the passband becomes effectively flat over a small
range of frequencies that correspond to discrete propagating modes in a inhomogeneously-
filled parallel plate waveguide [55] and is also found to hold for two-dimensional structures
(waveguides) as well [96].
Normalized propagation constant (k xa)
0.00 0.79 1.57 2.36 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
ky=0
ky=1/a
ky=1.5/a
ky=2/a
ky=2.5/a
Figure 2.6: Band structure (TMx) for lowest order mode of a periodic array of dielec-tric slabs as a function of off-axis propagation constant ky0 for filling fractionb/a=0.3545 and εr=8.9
Another check to determine if the code is calculating the correct propagation constant
is found by setting the filling fraction equal to some value x between zero (0) and one (1)
and obtaining a solution for the propagation constant. Subsequently, the filling fraction is
set equal to (1− x), the quantity (εr − 1) is replaced by (1− εr) in (2.21) and (2.25), and a
new solution is calculated. The solutions should be and are found to be exactly the same.
Asymptotic solutions are another effective way to check the solution. For small fill-
ing fractions, the propagation constant kx0 approaches k0√εeff where εeff is the effective
dielectric constant given by the volumetric average of the constitutive phases [24]
εeff =(1 + (εr − 1)
b
a
).
For a dielectric slab (εr=11) with a filling fraction of 0.1, the effective dielectric constant εeff
is estimated to be 2, corresponding to a propagation constant kx0 1.4k0. This approxima-
tion is only accurate to within 1% for relatively low normalized frequencies f0=fa/c < 0.1.
22
2.4 Formulation of Integral Equations
In this section, volume integral equations (IE) are derived from Maxwell’s equations
(2.1a)–(2.1d) and the boundary conditions, that when solved, yield propagation constants
from which the propagating mode structure can be determined. For the inhomogeneous
problems found throughout this work, integral equations are derived that incorporate equiv-
alent volume polarization currents.2 The integral equations are discretized and cast into a
matrix equation form through the use of the method of moments (MoM) numerical solu-
tion technique. The unknown is obtained from the solution of the resulting linear system.
For the problems solved in Chapters 2 and 3, the solution of the linear system returns the
eigenvalues (propagation constants) of the structure.
2.4.1 Derivation of Electric Field Integral Equation
In many scattering problems, the total electric field can be viewed as the sum of an
incident field Ei(r) due to radiation from a known source with the dielectric absent and a
scattered field E [J; r;V ] which is due to radiation by equivalent volume equivalent currents
J in a volume V
E(r) = E [J; r;V ] + Ei(r), (2.30)
in which the operator E [J; r;V ] can be expressed in terms of a Hertz potential [108] as
E [J; r;V ] = k20Π(r) +∇∇ · Π(r) (2.31a)
with
Π(r) =Z0
jk0
∫∫∫V
J(r′)e−jk0|r−r′|
4π|r − r′| dr′ (2.31b)
where k0 = ω√µ0ε0 is the free-space wavenumber of the background, ω is the frequency
of operation, Z0 =√µ0/ε0 is the intrinsic impedance of free-space, and V represents the
volume in which the sources reside. The field must satisfy
E [J; r;V ] + Ei(r) = E(r), r ∈ V. (2.32)
2In some homogeneous problems, integral equations are derived using equivalent surface currents reducingthe complexity and computational cost of determining the solution.
23
2.4.2 Equivalent Polarization Currents
The dielectric flux density D inside a dielectric immersed in a background medium of
free-space permittivity ε0 is equal to
D = ε0E + P (2.33)
where the polarization P is the electric dipole moment per unit volume. The polarization
current J is the time rate of change of the electric dipole moment per unit volume
J =∂
∂tP. (2.34)
Relating the polarization current to the difference between the induced electric flux density
and the electric flux density inside the dielectric through
J =∂
∂t(D − ε0E) (2.35)
yields the following expression for linear media
J =∂
∂t(εE − ε0E) . (2.36)
For time harmonic fields (ejωt), the equivalent polarization currents are related to the total
electric field by
J = jω (ε− ε0)E= jk0Y0 (εr − 1)E (2.37)
where Y0 = 1/Z0 is the intrinsic admittance of free-space.
TEx Case
Consider the representative dielectric slab (period not shown) in Figure 2.7 with period
a and dielectric insert width b. The slabs are excited by an incident time harmonic (ejωt)
plane wave with its electric field perpendicular to the plane of incidence (x-y plane) which
induces an electric current J with a z component only. Note that, since all quantities are
z invariant and because the induced electric current is z directed, ∇′ · J(r′) of (2.31a) is
zero. The total electric field everywhere is computed as the sum of the scattered field
produced by the equivalent induced electric current and the incident electric field given by
Ei = zE0e−jk0(x cosφ0+y sinφ0), which impinges on the slab in the k = (x cosφ0 + y sinφ0)
direction defined by an angle φi with respect to the x axis (φ0=0). This mode is denoted
24
y
xz
ba
E(H)
φ
Figure 2.7: Representative dielectric slab of width b with period a
as transverse electric to x, TEx, since for all angles of incidence the electric field does not
have a component in the x direction, the direction of stratification.
The dielectric material is replaced with equivalent volume currents using (2.37). For the
TEx case (assuming kz=0 without loss of generality),
Jz(x) = jk0Y0 (εr(x)− 1)Ez(x) (2.38)
where εr(x) is the permittivity function of the slabs within a unit cell given by
ε(x) =
εr, −b/2 < x < b/2;1, otherwise.
(2.39)
Since the dielectric material is periodic in x with period a, the resulting equivalent currents
must satisfy
Jz(x+ pa) = Jz(x)e−jkx0pa (2.40)
for a prescribed phase shift kx0a in the x direction. The scattered field is determined from
(2.31) by incorporating the periodic equivalent currents (2.40)
Esz(x) = −jk0Z0
b2∫
− b2
Jz(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′ (2.41)
where p is the Floquet mode index. Using Ez(x) = Esz(x) + E
iz(x), one can formulate the
25
following electric field integral equation (EFIE) to determine the equivalent currents
jk0Z0
b2∫
− b2
Jz(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′ +Jz(x)
jk0Y0(εr(x)− 1)= Ei
z(x). (2.42)
One-dimensional periodic free-space Green’s function The spatial form of the one-
dimensional periodic free-space Green’s function can be easily written in a more convenient
form through the use of the Poisson sum formula [70]. From Table I of [51],
∑p
f(p) =∑p
12jk
e−jk|x−pa|
=∑p
1a
[(2πpa
)2
− k2
]−1
e−j2πpax
= −1a
∑p
1[k2 −
(2πpa
)2]e−j 2πp
ax
(2.43)
where a is the unit cell width, k is the wavenumber of the medium, and p is the Floquet
index. Equation (2.43) becomes, upon substitution of a phase shift kx0 ,
∑p
e−jkx0pa1
2jke−jk|x−pa| = −1
a
∑p
e−jkxpx
k2 − k2xp
(2.44)
where
kxp =2πpa
+ kx0 .
Thus, (2.42) becomes upon substitution
−jk0Z0
a
b2∫
− b2
Jz(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ +Jz(x)
jk0Y0(εr(x)− 1)= Ei
z(x). (2.45)
TMx Case
Similarly for the the TMx case, the dielectric material is replaced with equivalent volume
currents. However, since the slabs are excited by an incident plane wave with its electric
field parallel to the plane of incidence, the induced electric current J can have both x and
y components. The total electric field everywhere is computed as the sum of the scattered
field produced by the equivalent induced electric current and the incident electric field given
by Ei = (z× k)E0e−jk0(x cosφi+y sinφi) which impinges on the slab in the k direction with φi
26
defined eariler. This mode is denoted as transverse magnetic to x, TMx, since for all angles
of incidence (assuming kz=0 without loss of generality), the magnetic field does not have a
component in the x direction.
For the TMx case, the dielectric material is replaced with equivalent volume currents
for both the x and y components
Jx(x) = jk0Y0(εr − 1)Ex(x) (2.46a)
Jy(x) = jk0Y0(εr − 1)Ey(x). (2.46b)
Using the periodic form of (2.46) in (2.31), the scattered field is found by integrating over
the induced current
Esx(x) = −jk0Z0
(1 +
1k20
∂2
∂x2
) b2∫
− b2
Jx(x′)∑p
e−jkx0pae−j
√k20−k2
y |x−x′−pa|
2j√k20 − k2
y
dx′
− kyZ0
k0
∂
∂x
b2∫
− b2
Jy(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′ (2.47a)
Esy(x) = −kyZ0
k0
∂
∂x
b2∫
− b2
Jx(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′
− jk0Z0
(1− k2
y
k20
) b2∫
− b2
Jy(x′)∑p
e−jkx0pae−j
√k20−k2
y |x−x′−pa|
2j√k20 − k2
y
dx′. (2.47b)
Using E(x) = Es(x) + Ei(x), one can formulate the following coupled EFIEs to determine
the equivalent currents
jk0Z0
(1 +
1k20
∂2
∂x2
) b2∫
− b2
Jx(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′
+kyZ0
k0
∂
∂x
b2∫
− b2
Jy(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′ +Jx(x)
jk0Y0(εr(x)− 1)= Ei
x(x)
(2.48a)
27
jk0Z0
(1− k2
y
k20
) b2∫
− b2
Jy(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′
+kyZ0
k0
∂
∂x
b2∫
− b2
Jx(x′)∑p
e−jkx0pae−j
√k20−k2
y|x−x′−pa|
2j√k20 − k2
y
dx′ +Jy(x)
jk0Y0(εr(x)− 1)= Ei
y(x).
(2.48b)
Using the Poisson sum formula (2.44) shown on page 26, and carrying out the derivatives
in (2.48) yields the coupled equations necessary to solve for the unknown equivalent currents
Jx and Jy,
− jk0Z0
a
(1− k2
xp
k20
) b2∫
− b2
Jx(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′
− kykxpZ0
jk0a
b2∫
− b2
Jy(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ +Jx(x)
jk0Y0(εr(x)− 1)= Ei
x(x) (2.49a)
− jk0Z0
a
(1− k2
y
k20
) b2∫
− b2
Jy(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′
− kykxpZ0
jk0a
b2∫
− b2
Jx(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ +Jy(x)
jk0Y0(εr(x)− 1)= Ei
y(x). (2.49b)
For a normally incident plane wave excitation, i.e. φi=0, the modes are degenerate (TEx
is equivalent to TMx).
2.4.3 Method of Moments Formulation
The first step in the method of moments (MoM) numerical method [34, 119] is to
discretize the geometry and approximate the unknown electric current in the dielectric
region with either subsectional or entire domain basis functions. The equivalent polarization
currents are expanded in a linear combination of these basis functions, and the equations
are tested in order to obtain an adequate number of equations to solve for the unknown
coefficients of the basis functions. In order to provide verification of the IE/MoM procedure,
computer codes have been written in Fortran to determine the propagating modes of a
periodic array of dielectric slabs.
28
Π Π
Π
x1
12
N
x2
xN∆x
Π3
(a) Piecewise constant basis functions
Λ 1Λ 2
ΛΝΛ 3
x2
x3
x1
xN∆x
(b) Piecewise linear basis functions
Figure 2.8: Piecewise constant and piecewise linear basis functions used in this work
TEx Case
Piecewise constant expansion / Piecewise constant testing The dielectric region
is discretized into N piecewise constant subsections, commonly referred to as pulse basis
functions, of width ∆x = b/N shown in Figure 2.8(a). The polarization current Jz is
approximated by a linear combination ofN piecewise constant basis functions with unknown
current coefficients Jzn located at the centers of the piecewise constant segments xn =
− b2 +∆x(n− 1
2),
Jz =N∑n=1
JznΠn(x) (2.50)
where
Πn(x) =
1, xn − ∆x
2 < x < xn + ∆x2 ;
0, otherwise.(2.51)
29
Inserting (2.50) into (2.42), yields
N∑n=1
Jzn
jk0Z0
a
xn+∆x2∫
xn−∆x2
∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ − Πn(x)jk0Y0 (εr − 1)
= Ei
z(x). (2.52)
Testing (2.52) with piecewise constant functions3 located at xm, and setting the excitation
to zero, yields the following equation
N∑n=1
Jzn
jk0Z0
a
xm+∆x2∫
xm−∆x2
xn+∆x2∫
xn−∆x2
∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ dx−xm+∆x
2∫xm−∆x
2
Πm(x)Πn(x)jk0Y0 (εr − 1)
dx
= 0, m = 1, 2, . . . , N (2.53)
which can be written in matrix form as[Zmn + δZ
] [Jzn
]=[0], (2.54)
where [Jzn ] is a column vector containing the unknown coefficients and [Zmn + δZ] is a
matrix whose elements are found by carrying out the integrations in (2.53). Specifically,
Zmn =jk0Z0∆2
x
a
∑p
sinc2(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
y − k2xp
(2.55a)
δZ = − δmn∆x
jk0Y0 (εr − 1). (2.55b)
The transforms of the piecewise constant testing and expansion functions, denoted Πm(kx)
and Πn(kx), are
Πm(kx) = ∆ sinc(kx∆2
)e−jkxxm (2.56a)
Πn(kx) = ∆ sinc(kx∆2
)ejkxxn . (2.56b)
Piecewise linear expansion / Piecewise linear testing Using piecewise linear func-
tions, commonly referred to as triangle basis functions, the dielectric region is discretized
into N subdomains of width ∆ = b/(N−1) that approximate the dielectric as shown in Fig-
ure 2.8(b). The polarization current Jz is approximated with a linear combination of these
N basis functions with unknown current coefficients Jzn centered at xn = − b2 +∆x(n−1),
Jz =N∑n=1
JznΛn(x) (2.57)
3When the expansion and testing functions are the same, the testing procedure is referred to as theGalerkin method [34, 83].
30
where
Λn(x) =
1− |x−xn|
∆x, xn −∆x < x < xn +∆x;
0, otherwise.(2.58)
Inserting the piecewise linear approximation for the current into (2.45) and testing the
resulting equation with piecewise linear basis functions, the elements of the impedance
matrix [Zmn + δZ] in (2.54) are found to be
Zm1 =jk0Z0∆2
x
a
∑p
(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−x1)
k20 − k2
y − k2xp
(2.59a)
Zmn =jk0Z0∆2
x
a
∑p
sinc4(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
y − k2xp
(2.59b)
ZmN =jk0Z0∆2
x
a
∑p
(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−xN )
k20 − k2
y − k2xp
(2.59c)
δZ = − δmn∆x
jk0Y0 (εr − 1). (2.59d)
The transforms of the piecewise linear testing and expansion basis functions, denoted
Λm(kx) and Λn(kx), are
Λm(kx) = ∆ sinc2(kx∆2
)e−jkxxm (2.60a)
Λn(kx) = ∆ sinc2(kx∆2
)ejkxxn . (2.60b)
The boundary condition at the edge of the dielectric insert requires continuity of the tan-
gential electric field. When using linear elements, this requires incorporating the two “half-
basis” functions, denoted Λ1(x) and ΛN (x), in Figure 2.8(b). The transforms of the two
“half-basis” functions in (2.59), Λ1 and ΛN , are
Λ1(kx) = ∆ejkxx1
(kx∆)2(1 + jkx∆− ejkx∆
)(2.61a)
ΛN (kx) = ∆ejkxxN
(kx∆)2(1− jkx∆− e−jkx∆
). (2.61b)
TMx Case
Piecewise linear expansion / Piecewise linear testing Care must be taken to cor-
rectly evaluate (2.49) because of the noticeable derivatives. The dielectric region is dis-
cretized into N subsections of width ∆ = b/(N − 1). The equivalent current J = xJx+ yJy
31
is expanded with a linear combination of N piecewise linear functions basis functions with
unknown current coefficients Jxn , Jyn centered at xn = − b2 +∆x(n− 1),
Jx =N∑n=1
JxnΛn(x) (2.62a)
Jy =N∑n=1
JynΛn(x) (2.62b)
with Λn(x) given in (2.58). Inserting (2.62) into (2.49) and testing the resulting equation
with piecewise linear functions yields
N∑n=1
Jxn
−jk0Z0
a
(1− k2
xp
k20
) xm+∆∫xm−∆
Λm(x)
xn+∆∫xn−∆
Λn(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ dx
+
xm+∆∫xm−∆
Λm(x)Λn(x)jk0Y0(εr(x)− 1)
dx
− Jyn
kykxpZ0
jk0a
xm+∆∫xm−∆
Λm(x)
xn+∆∫xn−∆
Λn(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ dx
= 0, m = 1, 2, . . . , N. (2.63a)
N∑n=1
Jxn
−kykxpZ0
jk0a
xm+∆∫xm−∆
Λm(x)
xn+∆∫xn−∆
Λn(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ dx
− Jyn
jk0Z0
a
(1− k2
y
k20
) xm+∆∫xm−∆
Λm(x)
xn+∆∫xn−∆
Λn(x′)∑p
e−jkxp(x−x′)
k20 − k2
y − k2xp
dx′ dx
+
xm+∆∫xm−∆
Λm(x)Λn(x)jk0Y0(εr(x)− 1)
dx
= 0, m = 1, 2, . . . , N. (2.63b)
Equation (2.63) can be written in matrix form asZxxmn Zxymn
Zyxmn Zyymn
+
δZxxmm 0
0 δZyymm
Jxn
Jyn
=
[0], (2.64)
32
where [Jxn Jyn ] is a column vector containing the unknown coefficients and [Zmn+ δZ] is a
matrix whose elements are found by carrying out the integrations in (2.63) yielding
Zm1xx =
jk0Z0∆2x
a
(1− k2
xp
k20
)∑p
(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−x1)
k20 − k2
xp
(2.65a)
Zmnxx =jk0Z0∆2
x
a
(1− k2
xp
k20
)∑p
sinc4(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
xp
(2.65b)
ZmNxx =jk0Z0∆2
x
a
(1− k2
xp
k20
)∑p
(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−xN )
k20 − k2
xp
(2.65c)
Zm1xy =
kykxpZ0∆2x
jk0a
∑p
(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−x1)
k20 − k2
xp
(2.65d)
Zmnxy =kykxpZ0∆2
x
jk0a
∑p
sinc4(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
xp
(2.65e)
ZmNxy =kykxpZ0∆2
x
jk0a
∑p
(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−xN )
k20 − k2
xp
(2.65f)
Zyx =Zxy (2.65g)
Zm1yy =
jk0Z0∆2x
a
(1− k2
y
k20
)∑p
(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−x1)
k20 − k2
xp
(2.65h)
Zmnyy =jk0Z0∆2
x
a
(1− k2
y
k20
)∑p
sinc4(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
xp
(2.65i)
ZmNyy =jk0Z0∆2
x
a
(1− k2
y
k20
)∑p
(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2 sinc2(kxp∆x
2
)e−jkxp(xm−xN )
k20 − k2
xp
(2.65j)
δZmmxx =− δmn∆x
jk0Y0 (εr − 1)(2.65k)
δZmmyy =− δmn∆y
jk0Y0 (εr − 1)(2.65l)
Piecewise constant expansion / Piecewise constant testing The solution can also
be carried out for the TMx case using piecewise constant expansion and testing functions.
The dielectric region is discretized into N subsections of width ∆ = b/N . The current
J = xJx + yJy is expanded in a linear combination of N piecewise constant basis functions
33
with unknown current coefficients Jxn , Jyn centered at xn = − b2 +∆x(n− 1
2 ),
Jx =N∑n=1
JxnΠn(x) (2.66a)
Jy =N∑n=1
JynΠn(x) (2.66b)
with Πn(x) given in (2.51). Inserting (2.66) into (2.49) and testing the resulting equation
with piecewise constant functions, the elements of the impedance matrix [Zmn + δZ] in
(2.64) are found to be
Zmnxx =jk0Z0∆2
x
a
(1− k2
xp
k20
)∑p
sinc2(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
xp
(2.67a)
Zmnxy =kykxpZ0∆2
x
jk0a
∑p
sinc2(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
xp
(2.67b)
Zmnyx =Zxy (2.67c)
Zmnyy =jk0Z0∆2
x
a
(1− k2
y
k20
)∑p
sinc2(kxp∆x
2
)e−jkxp(xm−xn)
k20 − k2
xp
(2.67d)
δZmmxx =− δmn∆x
jk0Y0 (εr − 1)(2.67e)
δZmmyy =− δmn∆y
jk0Y0 (εr − 1)(2.67f)
2.4.4 Matrix Solution
In order to accurately determine the eigenvalues of the impedance matrices given in
(2.55), (2.59), (2.65), and (2.67), a sufficient number of Floquet modes Np and subsectional
unknowns Nx must be included. A simple convergence criterion can be established to de-
termine when the computation of the individual impedance matrix elements has converged
adequately to its final value. A nontrivial solution for the fields requires the determinant
of the impedance matrix to be zero, which results in a characteristic equation. The eigen-
values (propagation constants) are obtained from the roots of this equation. For a lossless
structure, the propagation constants of a guided wave is a real number. However, in the
stopbands, the propagation constant is complex-valued.
The eigenvalues are calculated by minimizing a function of N variables subject to bounds
on the variables using a direct search algorithm. The algorithm uses the complex method
to find a local minimum point of a function of N variables and is based on function compar-
ison [139]. No derivative information is used in this technique. The minimization procedure
34
is carried out by fixing the frequency of operation and allowing the optimization code to
vary the propagation constant kx0 until a minimum is found. If more than one minimum
(eigenvalue) is found, implying that multiple modes are propagating in the structure, the
corresponding eigenvectors are analyzed to determine which eigenvalue corresponds to which
propagating mode. In the following tables, the propagation constant of the lowest order
mode that propagates is listed.
Table 2.1: Normalized TEx propagation constant kx0a for increasing number of Floquetmodes for f0=fa/c=1, εr=8.9, b=0.3545a, and Nx=40
kx0a (pulse) kx0a (triangle)
Np φ0=0 φ0=9.15 φ0=0 φ0=9.15
7 2.471 2.547 2.474 2.550
11 1.723 1.789 1.733 1.799
21 1.691 1.758 1.702 1.768
31 1.688 1.754 1.699 1.766
41 1.687 1.754 1.698 1.765
51 1.687 1.753 1.698 1.765
kx0a∗ 1.676 1.743 1.676 1.743
In Table 2.1, the normalized propagation constant kx0a for a normally incident plane
wave (φ0=0) and off-axis incident plane wave (φ0=9.15) for the TEx case is shown as
function of the number of Floquet modes (Np) and unknown basis set (pulse or triangle)
for the combination of normalized frequency f0=fa/c=1, 40 subsectional unknowns (Nx),
relative permittivity εr=8.9, and filling fraction b/a=0.3545. The value of φ0=9.15 cor-
responds to an off-axis propagation constant of ky0=1/a which is equal to the inverse of
the unit cell size. The off-axis propagation constant is set to this value because many au-
thors use this specific parameter value as a reference data set [38, 129]. The bold values
(error less than 1%) for Np=51 and Nx=40 listed in Tables 2.1–2.4 correspond to the same
solution and serve to represent a fixed parameter set to compare the relative convergence
as a function of number of unknowns and Floquet mode contributions. The correct value,
denoted kx0a∗ and listed at the bottom of Tables 2.1–2.4, is determined from a solution of
the exact eigenvalue equation derived earlier in the chapter.
Note in Table 2.1 that the necessary convergence occurs for both the pulse and triangle
basis cases for normal incidence and off-axis propagation as the number of Floquet modes
35
increase. The accuracy of the eigenvalue solution is limited by both Nx and Np. The
advantage of expanding the current in piecewise linear functions and testing with a similar
basis as opposed to piecewise constant expansion and testing functions would seem to be the
increase in convergence of the resulting series using the higher order bases. For the piecewise
constant expansion and testing functions, the Floquet series converges as O(p−4). However,
using piecewise linear expansion and testing functions, the convergence rate increases to
O(p−6). For even modest numbers of Floquet modes (Np=31), the solution has converged
to within 1% of the exact solution. Note the effect of the more complicated formulation
on the accuracy using the triangle bases is not immediately noticeable as a function of
increased Floquet contributions.
Table 2.2: Normalized TEx propagation constant kx0a for increasing number of unknownsfor f0=fa/c=1, εr=8.9, b=0.3545a, and Np=51
kx0a (pulse) kx0a (triangle)
Nx φ0=0 φ0=9.15 φ0=0 φ0=9.15
10 1.825 1.810 2.030 2.097
15 1.744 1.781 1.833 1.897
20 1.715 1.781 1.763 1.830
40 1.687 1.753 1.698 1.765
60 1.681 1.748 1.685 1.754
80 1.679 1.746 1.685 1.754
kx0a∗ 1.676 1.743 1.676 1.743
In Table 2.2, the normalized propagation constant for a normally incident plane wave
(φ0=0) and off-axis incident plane wave (φ0=9.15) for the TEx case is shown as func-
tion of the number of unknowns (Nx) and unknown basis set (pulse or triangle) for the
combination of normalized frequency f0=fa/c=1, 51 Floquet modes (Np), relative permit-
tivity εr=8.9, and filling fraction b/a=0.3545. Again, the results presented in Table 2.2
confirm that the necessary convergence occurs for both the pulse and triangle bases cases
for normal incidence and off-axis propagation as the number of unknowns increase. For
normalized frequency equal to unity (λ0=a), the electrical length of the dielectric insert
is 0.3545λ0/√εr=0.1188/λ0. Thus, only a few unknowns should be required to accurately
model the physical problem. For example, the solution obtained using only Nx=20 un-
knowns is within 5% of the exact solution. However, it is clear from the data that even
36
though the Floquet series may converge more quickly using higher order bases, the error
introduced to the final solution is dominated by the number of subsectional bases. Thus,
using the minimum number of bases if the most efficient solution to the problem.
The computational cost of computing each individual element in the impedance matrix
is proportional to Np. Although the implementation of pulse basis functions provides a
more slowly converging Floquet series than does the implementation using triangular bases,
it achieves a more accurate solution using fewer subsectional bases. The piecewise linear
functions are not only more expensive to compute than the simple pulse functions but are
also more prone to coding error. Since the total cost of computing the full impedance matrix
is proportional to N2xNp, incorporating fewer basis functions necessarily decreases this cost.
Tables 2.1 and 2.2 clearly show that the accuracy of computing the propagation constant is
more sensitive to the number of subsectional bases used in the approximation of the current
than the number of Floquet modes used to compute the Floquet series.
Table 2.3: Normalized TMx propagation constant kx0a for increasing number of Floquetmodes for f0=fa/c=1, εr=8.9, b=0.3545a, and Nx=40
kx0a (pulse) kx0a (triangle)
Np φ0=0 φ0=9.15 φ0=0 φ0=9.15
7 2.471 2.538 2.460 2.544
11 1.723 1.793 1.712 1.800
21 1.691 1.763 1.681 1.775
31 1.688 1.761 1.678 1.767
41 1.687 1.760 1.696 1.767
51 1.687 1.760 1.696 1.767
kx0a∗ 1.676 1.751 1.676 1.751
In Table 2.3, the normalized propagation constant for a normally incident plane wave
(φ0=0) and off-axis incident plane wave (φ0 = 9.15) for the TMx case is shown as func-
tion of the number of unknowns (Nx) and unknown basis set (pulse or triangle) for the
combination of normalized frequency f0=fa/c=1, 51 Floquet modes (Np), relative permit-
tivity εr=8.9, and filling fraction b/a=0.3545. As in the TEx case, the value φ0=9.15
corresponds to ky0=1/a, the inverse of the unit cell size. In Table 2.4, the normalized prop-
agation constant for a normally incident plane wave (φ0=0) and off-axis incident plane
wave (φ0=9.15) for the TEx case is shown as function of the number of Floquet modes
37
(Np) and unknown basis set (pulse or linear) for the combination of normalized frequency
f0=fa/c=1, 40 subsectional unknowns (Nx), relative permittivity εr=8.9, and filling frac-
tion b/a=0.3545.
Table 2.4: Normalized TMx propagation constant kx0a for increasing number of unknownsfor f0=fa/c=1, εr=8.9, b=0.3545a, and Np=51
kx0a (pulse) kx0a (triangle)
Nx φ0=0 φ0=9.15 φ0=0 φ0=9.15
10 1.824 1.898 2.029 2.093
15 1.744 1.817 1.834 1.901
20 1.715 1.788 1.765 1.833
40 1.687 1.760 1.696 1.767
60 1.681 1.755 1.685 1.760
80 1.679 1.753 1.685 1.751
kx0a∗ 1.676 1.751 1.676 1.751
Similar conclusions are drawn in Tables 2.3 and 2.4 for the TMx case that are ob-
served for the TEx case. The accuracy increases both by increasing the number of Floquet
mode contributions and by modeling the equivalent current more carefully by increasing the
number of unknowns. A simple pulse basis expansion serves to solve the problem quickly
and accurately. This observation will hold true as well for the two-dimensional structures
examined in Chapter 3.
In order to verify that the higher frequency solution of the propagation constants com-
puted using the method of moments technique and the plane wave expansion method are
correct, the band structure of a periodic array of dielectric slabs is calculated for a filling
fraction b/a=0.3545, relative dielectric constant εr=8.9, and off-axis4 propagation constant
ky0=1/a. Figure 2.9 shows the calculated band structure for both the TEx and TMx cases.
The moment method solution includes 40 subsectional bases and 51 Floquet modes. The
plane wave expansion method uses 63 Floquet modes. It is clear that the solutions ob-
tained by the two methods are in excellent agreement. For an off-axis propagation constant
of ky0 = 1/a and for relatively lower normalized frequency (f0 ≤ 0.4), the two modal so-
lutions (TEx, TMx) have different band gaps. At higher frequencies, the TEx and TMx
4By virtue of the fact that the TEx and TMx modes for are degenerate for normal incidence, an off-axispropagation case must be computed to be sure that the derivatives for the TMx case are formulated andimplemented correctly.
38
Normalized propagation constant (k xa)
0.00 0.79 1.57 2.36 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4 TM (FS)TE (FS)TM (MoM)TE (MoM)
Figure 2.9: Band structure (TEx, TMx) of a periodic array of dielectric slabs for b/a=0.3545, εr=8.9, and off-axis propagation constant ky0=1/a computed usingmethod of moments technique (MoM) with Np=51 and plane wave expansion(PWE) solution with Np=63
bands are indistinguishable for the prescribed parameter set.
2.5 Experimental Verification
Experimental measurements with micromachined dielectrics were performed to verify
that a similar band structure predicted theoretically using the method of moment solution or
plane wave expansion (eigenvalue) solution for plane wave excitation of periodic media exists
for microstrip excitation of a finite height periodic dielectric grating. Experimental and
computational evidence indicates that the high dielectric constant material used to support
the microstrip circuit effectively confines the field to near the region of the microstrip itself.
Consequently, the quasi-TEM microstrip excitation simply “sees” a periodically changing
dielectric constant – much like the periodic array of dielectric slabs excited using plane
wave excitation. To corroborate the frequency response obtained through measurements of
fabricated circuits, finite element simulations using the Ansoft High-Frequency Structure
Simulator (HFSS) were carried out to model the exact structure of interest. Additionally, an
equivalent model for a microstrip mounted over a periodic dielectric substrate (hi-Z, low-Z
filter) was simulated using the Agilent EEsof EDA Series IV microstrip Libra component.
A one-dimensional periodic dielectric substrate, or grating, shown in Figure 2.10 was
39
a
b
t
Figure 2.10: Microstrip mounted on 1-D periodic dielectric substrate with period a, fillingfraction b/a=0.5, and height t
fabricated with period a=6.35 mm (250 mil) and filling fraction equal to 0.5 using Rogers
Corporation RT/duroid having a dielectric constant of 10.2 and a thickness t of 0.635 mm
(25 mil). The measured response of the milled one-dimensional grating microstrip circuit is
shown in Figure 2.11. It is observed that the first stopband, defined by the 10 dB bandwidth
in the figure, is centered around 12 GHz with a second gap located between 21 and 26 GHz.
A third gap may or may not be visible around 36 GHz.
Frequency (GHz)
0 10 20 30 40
Inse
rtio
n L
oss
(dB
)
-50
-40
-30
-20
-10
0
Measured (1-D grating)Measured (2-D lattice)
Figure 2.11: Measured frequency response of a microstrip mounted on a 1-D grating and2-D lattice substrate
Concurrent to the circuit fabrication, the band structure for a one-dimensional periodic
array of dielectric slabs with the same geometrical and electrical parameters (a=6.35 mm,
40
b=3.175 mm, and εr=10.2) was determined. Three stopgaps were found to exist within the
measurement range of the HP8510 Network Analyzer, the first between 8.0 and 13.2 GHz,
a second between 18.7 and 26.2 GHz, and another between 31.2 and 37.3 GHz. These
results are in agreement with the measured response shown in Figure 2.11. The simulated
response of an HFSS simulation of the structure and the response of a Libra simulation of an
equivalent model are plotted in Figure 2.12. Due to the computational cost of implementing
the HFSS simulation of the realizable structure, only frequencies between 5 and 15 GHz
Frequency (GHz)
0 10 20 30 40
Inse
rtio
n L
oss
(dB
)
-30
-20
-10
0
HFSS simulationLibra model
Figure 2.12: Simulated frequency response of HFSS and Libra simulations of a microstripmounted on a 1-D grating
were analyzed. Note that both simulations reveal the first gap around 12 GHz and the
Libra model predicts the second and third gaps at 24 GHz and 36 GHz, respectively. The
gaps determined from the measurements and simulations of the finite structure reveal one
noteworthy difference between the gaps (stopbands) produced by a finite lattice (filter) and
the true gap (stopband) produced by an infinite lattice (filter), namely, that using a small
number of lattice periods (five) necessarily changes the shape and center frequency of the
response of the “filter.”
The encouraging results led to the question of how much material could be removed
while maintaining the response of the system, ultimately shedding light on the question of
how well the quasi-TEMmicrostrip mode in these periodic circuits can be modeled by simple
arrays of planar dielectric slabs. To this end, three substrates were designed that successively
added material to the one-dimensional grating to produce a finite one-dimensional grating,
a two-dimensional lattice of square air rods (holes), and finally, a two-dimensional lattice
41
of circular holes. Incorporating circular rods in the design instead of square ones has the
advantage of increased ease and speed of machining the substrates using simple milling
procedures. Each of the new substrates shown in Figure 2.13 was designed to have a period
equal to that of the one-dimensional periodic substrate shown in Figure 2.10. Consequently,
(a)
(b)
(c)
(d)
Figure 2.13: Microstrip mounted on (a) 1-D grating, (b) finite 1-D grating, (c) 2-D latticeof square holes, and (d) 2-D lattice of circular holes
the microstrip line would traverse effectively similar sections of dielectric material.
HFSS simulations were carried out for each of the three models for frequencies between
5 and 15 GHz to see the effect of the added material on the bandwidth and location of
the first stopgap. The results of the finite element simulations of the three “equivalent”
circuits are shown in Figure 2.14. Also shown is the response of a through line mounted
on a homogeneous substrate of same thickness. Note that, for all intents and purposes, the
responses of all of the periodic gratings are the same.
Finally, a two-dimensional periodic dielectric substrate, shown in Figure 2.15, was fabri-
cated with period a=6.35 mm (250 mil), cell-to-cell distance c=a/√2=4.49 mm (177 mil),
and circular hole diameter b=3.17 mm (125 mil) using RT/duroid having a dielectric con-
stant of 10.2 and a thickness t of 0.635 mm (25 mil). The physical dimensions of the two-
dimensional lattice were designed to produce the exact period a used in the one-dimensional
periodic substrate. The band structure was originally adapted from band plots presented
in [63] but has since been verified by solutions derived in Chapter 3. The frequency response
of the two-dimensional periodic substrate is also included in Figure 2.11.
42
Frequency (GHz)
5.0 7.5 10.0 12.5 15.0
Inse
rtio
n L
oss
(dB
)
-30
-25
-20
-15
-10
-5
0
Through line1-D gratingFinite 1-D grating2-D lattice of square holes2-D lattice of circular holes
Figure 2.14: Simulated frequency response of a microstrip mounted on a 1-D grating, finite1-D grating, 2-D lattice of square holes, and 2-D lattice of circular holes
a
t
d
c
Figure 2.15: Microstrip mounted on two-dimensional periodic dielectric substrate with pe-riod a, cell-to-cell distance c=a/
√2, hole diameter d, and height t
Noticeable differences in the frequency response of the one-dimensional grating and two-
dimensional lattice are observed in Figure 2.11. Not only are the bandwidths of the first two
stopbands of the two-dimensional substrate lattice narrower than the ones produced by the
one-dimensional grating, the center frequencies are shifted slightly. The HFSS simulations
shown in Figure 2.14 would seem to indicate that for microstrip excitation, the responses of
the one- and two-dimensional lattices shown in Figures 2.10 and 2.15 are indistinguishable.
This observation raises the question of how well HFSS can model the finite periodic dielectric
substrate.
43
2.6 Conclusions
The solution for the propagation of electromagnetic energy through a one-dimensional
periodic dielectric structure has been determined using two distinct solution techniques. A
Fourier series solution has been implemented by solving the one-dimensional (differential)
wave equation for periodic functions. Concurrently, a separate solution is obtained by de-
riving the integral equation solution for a one-dimensionally periodic media. The advantage
of using the plane wave expansion method over the IE/MoM solution is the computational
speed. The optimization procedure used in the IE/MoM solution requires the impedance
matrix to be filled for each iteration of the computation of the propagation constant. Even
using a highly efficient optimizer to quickly find the eigenvalue of interest, the process can
be somewhat time-consuming. For one-dimensional structures, this does not present a se-
rious problem. However, it will be clearly shown that for the two dimensional lattices of
Chapter 3, this restriction has grave consequences. The accuracy of the plane wave expan-
sion method increases with increased Floquet mode contributions whereas the accuracy of
the IE/MoM solution increases both by increasing the number of Floquet mode contribu-
tions and by modeling the equivalent current more carefully by increasing the number of
unknowns. It is observed that, for the IE/MoM solution, a simple pulse basis expansion
serves to solve the problem quickly and accurately. This observation will hold true as well
for the two-dimensional structures examined in Chapter 3.
Experimental verification of the one-dimensional band gaps is accomplished by determin-
ing the filtering effect of mounting a microstrip over a one-dimensional periodic substrate.
The quasi-TEM fields of the microstrip excitation for the periodic substrate are modeled as
plane waves incident on a one-dimensionally periodic medium. The band structure of the
resulting microwave circuit is also modeled effectively using a simple hi-Z, low-Z filter and
is similar to the band structure determined from the plane wave solution.
44
CHAPTER 3
Two-Dimensional Periodic Dielectric Structures
3.1 Introduction
As opposed to one-dimensional structures, where many applications are found at the
higher quasi-optical and optical frequencies [1, 2], research into applications involving two-
dimensional periodic structures is fruitful at microwave and lower frequencies. In fact, an
entire class of artificial surfaces and/or materials termed frequency selective surfaces (FSS)
incorporates the use of period to provide unique spectral characteristics in applications rang-
ing from radar cross section reduction to antenna array beam-forming. Two-dimensional
dielectric lattices are actively being incorporated into microwave circuit and antenna de-
signs, high-Q filters and resonators, waveguides, and novel materials/surfaces. Although
serious attention has been focused by the electromagnetics community on traditional fre-
quency selective surfaces, new applications have recently emerged for more general two-
dimensional periodic structures and are challenging traditional concepts and views about
electromagnetic propagation in periodic media [3, 4, 5].
This chapter extends the solutions and techniques developed for one-dimensional period
in Chapter 2 to structures that have periodicities in two directions. The phase constants for
electromagnetic waves propagating in a two-dimensional lattice for both dielectric rods in
an air background and air columns immersed in a dielectric media are found explicitly using
two similar but decidedly different techniques. The first solution, detailed in Section 3.2,
incorporates the use of a Fourier series representation for the periodic field and is determined
by solving the differential equation (two-dimensional wave equation) for periodic media.
The second solution is found by deriving an integral equation for the periodic field and
numerically solving the resulting linear system and is presented in Section 3.3.
45
In light of the complicated nature of two-dimensional periodic media, effective medium
theory (EMT) is applied to the lattice which reduces the problem to that of solving for
the propagation in an equivalent one-dimensional array with an effective permittivity. In
Section 3.4, the theory of EMT is outlined briefly and its advantages and limitations are
discussed. One of the significant disadvantages of EMT is its limitation to electrically small
lattice elements. A number of representative structures are used in the chapter to show the
behavior of electromagnetic fields in two-dimensional periodic media and observations about
the salient features of each are discussed. The particular application of a two-dimensional
periodic dielectric structure in parallel-plate mode reduction in conductor-backed slots is
presented extensively in Appendix D. For the structures shown in this chapter, only in-
plane propagation is considered. This is significant because complete band gaps1 only exist
for in-plane propagation.
1A complete band gap (for a given polarization) is defined as one where over a range of frequencies(band), electromagnetic energy may not propagate in any direction. Some authors define a complete bandgap as one where for any electromagnetic wave polarization, energy may not propagate in any direction.
46
3.2 Plane Wave Expansion Method
3.2.1 Analytical Techniques
For two-dimensional periodic problems, the solution of the exact eigenvalue equation can
be obtained through the use of a double Fourier series. A representative two-dimensional
E(H)
φ
ba
y
xz
Figure 3.1: Cross-sectional view of two-dimensional array of dielectric rods of diameter b ina periodic square lattice with period a
periodic array of dielectric rods of diameter b in a periodic square lattice2 with period a is
illustrated in cross section in Figure 3.1. If the electric field has only an axial (or z -directed)
component, the mode is transverse magnetic to z and is denoted TMz. If the electric field
has only components in the transverse x-y plane, the mode is transverse electric to z and
is denoted TEz.
TMz Case
The electric field is expanded as a periodic function with period a in the x -direction and
prescribed propagation constant kx0 and with period c in the y-direction and prescribed
propagation constant ky0 and is given by
E(x, y) = zEz(x, y) = zEp(x, y) e−jkx0x e−jky0y (3.1)
2Square lattices are used in this chapter to simplify the notation and understanding. Extensions of thesolutions to other two-dimensional lattices depicted in Figure B.3 of Appendix B is straightforward.
47
where Ep(x, y) is the periodic electric field that propagates only in the xy-plane. Again,
applying the operator (∇2xy + k2) to the electric field in (3.1) yields an equation similar
to (2.15) with εr(x) replaced with εr(x, y). If the rods are assumed to be infinite in the z
direction(∂∂z = 0
), then the wave equation simplifies to
− ∂2
∂x2Ez(x, y)− ∂2
∂y2Ez(x, y) = k2
0εr(x, y)Ez(x, y) (3.2)
If the periodic field, expanded as in a double Fourier series in x and y with unknown
coefficients anq which serve to represent the dependence on z,
Ep(x, y) =∑n
∑q
anq e−j 2πn
ax e−j
2πqcy, (3.3)
and the periodic dielectric rods, expanded in another double Fourier series with coefficients
bmp,
εr(x, y) =∑m
∑p
bmp e−j 2πm
ax e−j
2πpay, (3.4)
are substituted into (3.2) and the algebraic operations are carried out, then
∑n
∑q
[(2πna
+ kx0
)2
+(2πqc
+ ky0
)2]anq e
−j 2πnax e−j
2πqcy =
k20
∑m
∑p
bmp e−j 2πm
ax e−j
2πpay∑n
∑q
anq e−j 2πn
ax e−j
2πqcy. (3.5)
In order to determine the coefficients anq and bmp, (3.5) is multiplied by orthogonal functions
in x and y and integrated over the unit cell. This integration produces a Kronecker delta
function for specific indices in the x and y directions and
∑n
∑q
[(2πna
+ kx0
)2
+(2πqc
+ ky0
)2]anqδ
(2πlc
− 2πqc
)δ
(2πka
− 2πna
)=
k20
∑m
∑p
∑n
∑q
anqbmpδ
(2πlc
− 2πpc
− 2πqc
)δ
(2πka
− 2πma
− 2πpa
). (3.6)
The convolution in (3.6) can be cast, albeit not as easily as (2.20), into the following general
matrix form
anq
[(2πna
+ kx0
)2
+(2πqc
+ ky0
)2]= k2
0
∑m
∑p
anqbn−m,q−p (3.7)
where for the square dielectric rod illustrated in Figure 3.2(a),
48
a
d
b
c
(a)
a
b
(b)
Figure 3.2: Unit cell dimensions for (a) rectangular dielectric rods and (b) circular dielectricrods where for the rectangular rod, a and c are the unit cell sizes, b and d arethe element edge lengths, and for the circular rod, a is the unit cell size and bis the diameter
bn−m,q−p =1ac
b/2∫−b/2
d/2∫−d/2
(εr − 1) e−j2π(n−m)
ax e−j
2π(q−p)a
y dx dy
+1ac
a/2∫−a/2
c/2∫−c/2
(1) e−j2π(n−m)
ax e−j
2π(q−p)a
y dx dy
=bd
ac(εr − 1) sinc
π(n−m)ba
sincπ(q − p)d
c+ δn−m,q−p (3.8)
and for the circular dielectric rod illustrated in Figure 3.2(b),
bn−m,q−p =2πa2
b/2∫0
(εr − 1)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
+2πa2
a∫0
(1)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
=πb2
a2(εr − 1)
2J1
(√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2)
√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2+ δn−m,q−p. (3.9)
In Figure 3.2(a), a and c are the unit cell sizes, b and d are the square element edge lengths.
In Figure 3.2(b), a is the unit cell size and b is the diameter of the circular element. With
a little algebraic work, the coefficients bn−m,q−p in (3.8) and (3.9) can be manipulated to
agree with similar expressions in [107] and [73].
49
If the dielectric rods are replaced by air columns, the coefficients bn−m,q−p become
bn−m,q−p =1ac
a/2∫−a/2
c/2∫−c/2
(εr) e−j2π(n−m)
ax e−j
2π(q−p)a
y dx dy
− 1ac
b/2∫−b/2
d/2∫−d/2
(εr − 1) e−j2π(n−m)
ax e−j
2π(q−p)a
y dx dy
=εrδn−m,q−p − (εr − 1)bd
acsinc
π(n−m)ba
sincπ(q − p)d
c(3.10)
for the rectangular air column shown in Figure 3.3(a), and
a
d
b
c
(a)
a
b
(b)
Figure 3.3: Unit cell dimensions for (a) rectangular dielectric air columns and (b) circularair columns where for the rectangular column, a and c are the unit cell sizes, band d are the element edge lengths, and for the circular column, a is the unitcell size and b is the diameter
bn−m,q−p =2πa2
a∫0
(εr)J0
√(
2π(n−m)ra
)2
+(2π(q − p)r
a
)2 r dr
− 2πa2
b/2∫0
(εr − 1)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
=εrδn−m,q−p − (εr − 1)πb2
a2
2J1
(√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2)
√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2(3.11)
for the circular air column shown in Figure 3.3(b).
A generalized linear eigensystem problem is represented by Ax = λBx where A and B
are n×n matrices. The value λ is an eigenvalue and x = 0 is the corresponding eigenvector.
50
The propagating modes in the TMz case are solutions of the generalized linear eigensystem
in (3.7).
TEz Case
For the TEz case, the electric field is replaced by the magnetic field in the wave equation
∇×
1εr(x, y)
∇× zHz(x, y)+ k2
0zHz(x, y) = 0 (3.12)
where εr(x) in (2.23) has been replaced with εr(x, y). The same care must be taken to
apply the wave equation operator to the magnetic field in the two dimensional case that
was taken to apply it to the magnetic field in the one dimensional case. The resulting
equation resembles (2.24)
1εr(x, y)
z · ∇ ×∇× H(x, y) + z · ∇
1εr(x, y)
×∇× H(x, y) = −k2
0Hz(x, y). (3.13)
Expanding the periodic magnetic field in a double Fourier series,
Hp(x) =∑n
∑q
anq e−j 2πn
ax e−j
2πqcy, (3.14)
and the inverse of the dielectric function in another double Fourier series as
1εr(x, y)
=∑m
∑p
bmp e−j 2πm
ax e−j
2πpay, (3.15)
substituting the expansions into (3.13), and carrying out the curl and gradient operations
yields
∑m
∑p
bmp e−j 2πm
ax e−j
2πpay
− ∂2
∂x2
∑n
∑q
anq e−j 2πn
ax e−j
2πqcy e−jkx0xe−jky0y
− ∂2
∂y2
∑n
∑q
anq e−j 2πn
ax e−j
2πqcy e−jkx0x e−jky0y
− ∂
∂x
∑m
∑p
bmp e−j 2πm
ax e−j
2πpay ∂
∂x
∑n
∑q
anq e−j 2πn
ax e−j
2πqcy e−jkx0x e−jky0y
− ∂
∂y
∑m
∑p
bmp e−j 2πm
ax e−j
2πpay ∂
∂y
∑n
∑q
anq e−j 2πn
ax e−j
2πqcy e−jkx0x e−jky0y
= −k20
∑n
∑q
anq e−j 2πn
ax e−j
2πqcy e−jkx0x e−jky0y. (3.16)
Simplifying (3.16) by carrying out the derivatives and combining like series, integrating the
result over one period, and simplifying the resulting expression yields an equation similar
to (2.28)
51
∑m
∑p
anqbn−m,q−p
[(2πna
+ kx0
)2
+(2πqc
+ ky0
)2
− 2π(n−m)a
(2πna
+ kx0
)− 2π(q − p)
c
(2πqc
+ ky0
)]= −k2
0anq (3.17)
where for rectangular rods,
bn−m,q−p =1ac
b/2∫−b/2
d/2∫−d/2
(1εr
− 1)e−j
2π(n−m)a
x e−j2π(q−p)
ay dx dy
+1ac
a/2∫−a/2
c/2∫−c/2
(1)e−j2π(n−m)
ax e−j
2π(q−p)a
y dx dy
=bd
ac
(1εr
− 1)sinc
π(n−m)ba
sincπ(q − p)d
c+ δn−m,q−p (3.18)
and for circular rods,
bn−m,q−p =2πa2
b∫0
(1εr
− 1)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
+2πa2
a∫0
(1)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
=πb2
a2
(1εr
− 1) 2J1
(√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2)
√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2+ δn−m,q−p (3.19)
Again if the dielectric rods are replaced by air columns as before, the coefficients bn−m,q−p
become
bn−m,q−p =1ac
a/2∫−a/2
c/2∫−c/2
(1εr
)e−j
2π(n−m)a
x e−j2π(q−p)
ay dx dy
− 1ac
b/2∫−b/2
d/2∫−d/2
(1εr
− 1)e−j
2π(n−m)a
x e−j2π(q−p)
ay dx dy
=1εrδn−m,q−p −
(1εr
− 1)bd
acsinc
π(n−m)ba
sincπ(q − p)d
c, (3.20)
52
for the rectangular air column shown in Figure 3.3(a), and
bn−m,q−p =2πa2
a∫0
(1εr
)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
− 2πa2
b/2∫0
(1εr
− 1)J0
√(
2π(n −m)ra
)2
+(2π(q − p)r
a
)2 r dr
=1εrδn−m,q−p −
(1εr
− 1)πb2
a2
2J1
(√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2)
√(2π(n−m)b
a
)2+(
2π(q−p)ba
)2, (3.21)
for the circular air column in Figure 3.3(b). The explicit expressions for the coefficients
bn−m,q−p in (3.10), (3.11), (3.18), (3.19), (3.20), and (3.21) have not been found in the
literature.
An ordinary linear eigensystem problem is represented by the equation Ax = λx where
A denotes an n × n matrix. The propagating modes in the TEx case are solutions of the
ordinary eigensystem problem in (3.17).
3.2.2 Matrix Solution of the Eigensystem
The resulting eigenvalues of the matrices in (3.7) and (3.17) are the squares of the
frequencies of the propagating modes in the structure. The solutions of the frequencies
of the propagating modes in the structure are found for specific values of kx0a, ky0a ∈[0, 2π]. The full band structure3 (TMx) for a normally incident mode in a periodic array of
dielectric rods with b/a=0.3545 and εr=8.9 is shown in Figure 3.4. The values for the filling
fraction and the relative dielectric constant are obtained from a two-dimensional structure
in [64, 129]. The Brillouin zone for a square lattice is also included in the figure where
the Γ, X, and M points are defined in Appendix B. A complete TM band gap exists over
a significant range of frequencies. Certainly, larger incomplete bands exist (particularly
in the Γ–M and Γ–X directions) where propagation is allowed for specific directions and
these gaps are effectively used in designs that do not need total omni-directional stopbands.
Notice that there exists no complete TE band gap. In fact, only a small TE gap exists in
any direction (note the small gap at the X point in Figure 3.4) for this geometrical and
electrical configuration.3A full band structure is one where the propagation constant is determined along the edges of the Brillouin
zone. Theoretically, one must determine the propagation constant for all of the unique combinations of kx0
and ky0 . Fortunately, physics dictates that the band gaps are found at the edges of the Brillouin zone (BZ).Thus, by sampling the edge of the BZ, one can quickly determine the requisite full band structure.
53
0.00 3.14 6.28 9.42
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.25
0.50
0.75
1.00
TMz
TEz modes
modes
band gapTM
ΓX
M
Γ βx (βy=0) X βy (βx=π/a) M βx=βy Γ
Normalized propagation constant ( βa)
Figure 3.4: Band structure (TMz) for a normally incident mode in a periodic array ofdielectric rods with b/a=0.3545 and εr=8.9
Extensive research has concluded that large complete TM band gaps are most likely to be
found using isolated regions of high dielectric constant material imbedded in a background of
lower dielectric constant, whereas for TE band gaps, significant bands are found for isolated
regions of lower dielectric constant immersed in a higher dielectric background [38, 129].
The latter conclusion can be seen in Figure 3.5 where the full band structure (TEx) for a
normally incident mode is shown for the periodic array of air columns with b=0.6455a and
background dielectric constant εr=8.9. For this structure, which is complementary to the
structure in Figure 3.4, no complete TM gap can be found. Small TM gaps been seen at
the X and M points in the lattice but are too small to be of much value.
If the structures are viewed from a connectivity point of view, new insights can be
gleaned from the preceding figures. The air cross of Figure 3.6(a) is equivalent to the
dielectric square of Figure 3.2(a). Thus, regions of air that are “connected” to other cells
by thin veins of air are useful for TM band gaps. The complementary dielectric cross of
Figure 3.6(b) is equivalent to the air square of Figure 3.3(a). For useful TE band gaps,
regions of dielectric should be connected to other cells by thin veins of dielectric. Either
interpretation reveals that the electric field maintains a preference to distribute itself in
certain ways depending on the polarization.
A simple check of the correctness of the results obtained from the two-dimensional pe-
riodic structure is to allow the filling fraction in one lattice direction to increase until an
54
0.00 3.14 6.28 9.42
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.25
0.50
TMz modes
TEz modes
Γ βx (βy=0) X βy (βx=π/a) M βx=βy Γ
Normalized propagation constant ( βa)
band gapTE
ΓX
M
Figure 3.5: Band structure (TEz) for a normally incident mode in a periodic array of aircolumns with b/a=0.6455 and εr=8.9
(a) (b)
Figure 3.6: Equivalent structures for Figure 3.2(a) and Figure 3.3(a)
equivalent one-dimensionally periodic structure is obtained. As the unit cell geometry ap-
proaches the one-dimensional periodic structure of equal filling fraction, the propagation
constant should approach the value of the one-dimensional propagation constant (ky=0). In
Figure 3.7 the normalized TMz propagation constant kx0a is shown as a function of d/c for
normalized frequency f0=fa/c=0.5, relative dielectric constant εr=10.2, normal incidence,
and filling fraction b/a=0.2. Comparing the results with the normalized propagation con-
stant obtained from the equivalent one-dimensional structure, k1Dx0a, one can conclude that
indeed the solutions are the same. Similar results can be obtained for the normalized TEz
propagation constant kx0a as a function of d/c.
55
k =2.85 k =2.58 k =2.33k =2.43 k =2.210k =2.24x xxxxx
Ez
x
y
d/c=0.2 d/c=0.95d/c=0.8d/c=0.6d/c=0.4 d/c=1.0
b/a=0.2
k =2.204x1D
Figure 3.7: Normalized TMz propagation constant kx0a as a function of d/c for f0=fa/c=0.5, εr=10.2, ky=0, and b/a=0.2
3.3 Formulation of Integral Equations
Great physical insight is often obtained through the necessary theoretical formulation
of a problem from basic principles. This is particularly true when deriving two- and three-
dimensional integral equations (IE) that when solved, yield the unknowns of interest. Ap-
plying the method of moments (MoM) numerical solution technique, where the physics of
the problem is often intimately related to the formulation itself, also provides intuition
about the quantities being studied.
3.3.1 Derivation of Electric Field Integral Equation
Similarly to the derivation of the one-dimensional case, the dielectric rods are replaced
by equivalent (polarization) volume currents and the total field is determined as the sum
of the incident field produced by a known source with the dielectric absent and a scattered
field contributed by the equivalent currents induced in the periodic scatterers.
TMz Case
Consider the representative dielectric rod (period not shown) of diameter b in a periodic
square lattice with period a in Figure 3.8. The rods are excited by an incident time harmonic
(ejωt) plane wave propagating in a direction normal to the z axis (in-plane propagation)
with its electric field parallel to the axis (TMz) which induces an electric current J = Jz z
with a z component only. Note that, for uniformity in the z direction, ∇ · J of (2.31)
is zero. The total electric field everywhere is computed as the sum of the scattered field
produced by the equivalent induced electric current and the incident electric field given
by Ei(x, y) = zE0e−jk0(x cosφ0+y sinφ0), which impinges upon the rod along a ray in the
56
E(H)
y
xz
φ
b
a
Figure 3.8: Representative dielectric rod of diameter b in a periodic square lattice withperiod a
k = (x cosφ0+y sinφ0) direction defined by an angle φ0 with respect to the x axis (φ0 = 0).
The first step is to replace the dielectric material with equivalent volume currents using
(2.37). If we assume for simplicity that the two-dimensional periodic array in Figure 3.1
is composed of dielectric rectangular rods of cross section b×d in a periodic square lattice
with cross section a×c, then for the TMz case,
Jz(x, y) = jk0Y0 (εr(x, y)− 1)Ez(x, y) (3.22)
where εr(x, y) is the permittivity function of the rods defined by
εr(x, y) =
εr, −b/2 < x < b/2,−d/2 < y < d/2;1, otherwise.
(3.23)
Since the dielectric material is periodic in x with period a and in y with period c, the
resulting equivalent currents must satisfy
Jz(x+ pa, y + qc) = Jz(x, y)e−jkx0pae−jky0qc (3.24)
for phase shifts kx0a in the x direction and ky0c in the y direction. Using (2.31), the
scattered field is determined from the periodic equivalent currents (3.24) to be
Esz(x, y) = −jk0Z0
b2∫
− b2
d2∫
− d2
Jz(x′, y′)Gp(x, y;x′, y′) dx′ dy′ (3.25)
57
where Gp(x, y;x′, y′) is the two-dimensional periodic free-space Green’s function (PFSGF)
given by
Gp(x, y;x′, y′) =∑p,q
e−jkx0pa e−jky0qc14jH
(2)0
(k0
√(x− x′ − pa)2 + (y − y′ − qc)2
)
(3.26)
and p, q are the Floquet mode indices for propagation in the x and y directions, re-
spectively. Using Ez(x, y) = Esz(x, y) + Ei
z(x, y), one can formulate the following EFIE to
determine the equivalent currents
jk0Z0
b2∫
− b2
d2∫
− d2
Jz(x′, y′)Gp(x, y;x′, y′) dx′ dy′ +Jz(x, y)
jk0Y0 (εr(x, y)− 1)= Ei
z(x, y). (3.27)
Two-dimensional periodic free-space Green’s function The extension of the one-
dimensional periodic free-space Green’s function to a periodic two-dimensional periodic
free-space Green’s function is straight forward. The spatial form of the two-dimensional
free-space periodic Green’s function can also be easily transformed into its spectral form
through the use of the Poisson sum formula [70]. From Table I of [51],
∑p,q
f(p, q) =∑p,q
14jH
(2)0
(k
√(x− pa)2 + (y − qc)2
)
=1a
∑p,q
1
2j
√k2 −
(2πpa
)2e−j
√k2−( 2πp
a )2|y−qc|e−j2πp
ax.
(3.28)
Again, using Table I of [51],
1a
∑p,q
1
2j
√k2 −
(2πpa
)2e−j
√k2−( 2πp
a )2|y−qc|e−j2πpax
= − 1ac
∑p
1[k2 −
(2πpa
)2 −(
2πqc
)2]e−j 2πp
axe−j
2πqcy (3.29)
where a is the unit cell width in the x-direction, b is the width of the dielectric region in
the x-direction, c is the unit cell depth in the y-direction, d is the depth of the dielectric
region in the y-direction, k is the wavenumber of the medium, and p, q are the Floquet
indices. Equation (3.29) becomes upon substitution of a phase shift k = kx0 x+ ky0 y
∑p,q
14jH
(2)0
(k
√(x− pa)2 + (y − qc)2
)e−jkx0pae−jky0qc = − 1
ac
∑p,q
e−jkxpxe−jkyqy
k2 − k2xp
− k2yq
(3.30)
58
where
kxp =2πpa
+ kx0
kyq =2πqc
+ ky0 .
Thus, (3.27) becomes upon substitution
− jk0Z0
ac
b2∫
− b2
d2∫
− d2
Jz(x′, y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′ +Jz(x, y)
jk0Y0 (εr(x, y)− 1)
= Eiz(x, y). (3.31)
TEz Case
Similarly for the the TEz case, we replace the dielectric material with equivalent vol-
ume currents. However, note that since the rods are excited by an incident plane wave
propagating in a direction normal to the z axis (in-plane propagation) with its magnetic
field parallel to the axis (TEz), the induced electric current J has only components in the
transverse (x,y) plane.
In this case, the dielectric material is replaced with equivalent volume currents for both
the x and y components
Jx(x, y) = jk0Y0 (εr(x, y)− 1)Ex(x, y) (3.32)
Jy(x, y) = jk0Y0 (εr(x, y)− 1)Ey(x, y). (3.33)
Substituting the periodic polarization currents into (2.31), the scattered field can be deter-
mined to be
Esx(x, y) = −jk0Z0
(1 +
1k20
∂2
∂x2
) b2∫
− b2
d2∫
− d2
Jx(x′, y′) Gp(x, y;x′, y′) dx′ dy′
− Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Jy(x′, y′) Gp(x, y;x′, y′) dx′ dy′ (3.34a)
59
Esy(x, y) = − Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Jx(x′, y′) Gp(x, y;x′, y′) dx′ dy′
− jk0Z0
(1 +
1k20
∂2
∂y2
) b2∫
− b2
d2∫
− d2
Jy(x′, y′) Gp(x, y;x′, y′) dx′ dy′. (3.34b)
Using (2.32), one can formulate the following coupled EFIEs to determine the equivalent
currents
jk0Z0
(1 +
1k20
∂2
∂x2
) b2∫
− b2
d2∫
− d2
Jx(x′, y′) Gp(x, y;x′, y′) dx′ dy′
+Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Jy(x′, y′) Gp(x, y;x′, y′) dx′ dy′ +Jx(x, y)
jk0Y0 (εr(x, y)− 1)= Ei
x(x, y)
(3.35a)
Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Jx(x′, y′) Gp(x, y;x′, y′) dx′ dy′ +Jy(x, y)
jk0Y0 (εr(x, y) − 1)
+ jk0Z0
(1 +
1k20
∂2
∂y2
) b2∫
− b2
d2∫
− d2
Jy(x′, y′) Gp(x, y;x′, y′) dx′ dy′ = Eiy(x, y). (3.35b)
Using the Poisson sum formula (3.30) shown on page 58, (3.35) becomes
jk0Z0
(1 +
1k20
∂2
∂x2
) b2∫
− b2
d2∫
− d2
Jx(x′, y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
+Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Jy(x′, y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′ +Jx(x, y)
jk0Y0 (εr(x, y) − 1)
= Eix(x, y) (3.36a)
60
Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Jx(x′, y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′ +Jy(x, y)
jk0Y0 (εr(x, y)− 1)
+ jk0Z0
(1 +
1k20
∂2
∂y2
) b2∫
− b2
d2∫
− d2
Jy(x′, y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
= Eiy(x, y). (3.36b)
3.3.2 Method of Moments Formulation
For two-dimensional problems, the choice of discretization can have a significant effect
on the reduction of the integral equations into matrix equations. After discretizing the
geometry and expanding the unknown equivalent electric current across the dielectric region
with a linear combination of subsectional or entire domain basis functions, the equations
are tested in order to obtain an adequate number of equations to solve for the unknown
coefficients of the basis functions. In order to provide verification of the IE/MoM procedure,
computer codes have been written in Fortran to determine the propagating modes of a
periodic array of dielectric slabs.
TMz Case
Piecewise constant expansion / Piecewise constant testing For piecewise constant
functions, we discretize the dielectric region into N=NxNy subsections of width ∆x = b/Nx
and ∆y = d/Ny, respectively, to approximate the original surface and represent Jz by a
linear combination of N piecewise constant basis functions Πn(x, y) with unknown current
coefficients Jn
Jz(x, y) =N∑n=1
JnΠn(x, y) =N∑n=1
JnΠn(x)Πn(y) (3.37)
where
Πn(x) =
1, xn − ∆x
2 < x < xn + ∆x2 ;
0, otherwise;(3.38)
and
Πn(y) =
1, yn − ∆y
2 < y < yn +∆y
2 ;
0, otherwise;(3.39)
61
and xn = − b2 +∆x(n− 1
2) and yn = −d2 +∆y(n− 1
2).
Substituting the currents into (3.31) yields
N∑n=1
Jn
jk0Z0
ac
xn+∆x2∫
xn−∆x2
yn+∆y2∫
yn−∆y2
∑p,q
e−jkxp(x−x′)e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dy′ dx′ − Πn(x)Πn(y)jk0Y0 (εr(x, y)− 1)
= 0. (3.40)
Testing (3.40) with piecewise constant functions centered at (xm, ym) yields the following
equation
N∑n=1
Jn
jk0Z0
ac
xm+∆x2∫
xm−∆x2
ym+∆y2∫
ym−∆y2
xn+∆x2∫
xn−∆x2
yn+∆y2∫
yn−∆y2
∑p,q
e−jkxp(x−x′)e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dy′ dx′ dy dx
−xm+∆x
2∫xm−∆x
2
ym+∆y2∫
ym−∆y2
Πn(x)Πn(y)jk0Y0 (εr(x, y) − 1)
dx dy
= 0, m = 1, 2, . . . , N. (3.41)
By carrying out the integrations in (3.41) and simplifying the resulting expressions, the
impedance matrix elements can be determined explicitly and are listed in Appendix C. The
resulting matrix equation is written in a form similar to (2.54).
The transforms of the testing and expansion functions, denoted Πm(kx) and Πn(kx),
are given in (2.56). The transforms of the expansion and testing functions, denoted Πm(ky)
and Πn(ky), respectively, are
Πm(ky) = ∆ sinc(ky∆2
)e−jkyym (3.42)
Πn(ky) = ∆ sinc(ky∆2
)ejkyyn . (3.43)
Piecewise linear expansion / Piecewise linear testing For piecewise linear functions,
we discretize the dielectric region into N=NxNy subsections of width ∆x = b/(Nx− 1) and
∆y = d/(Ny−1), respectively, that approximate the original surface and approximate Jz by
a linear combination of N piecewise linear basis functions Λn(x, y) with unknown current
coefficients Jn
Jz(x, y) =N∑n=1
JnΛn(x, y) =N∑n=1
JnΛn(x)Λn(y) (3.44)
62
where
Λn(x) =
1− |x−xn|
∆x, xn −∆x < x < xn +∆x;
0, otherwise,(3.45)
and
Λn(y) =
1− |y−yn|
∆y, yn −∆y < y < yn +∆y;
0, otherwise,(3.46)
where xn = − b2 +∆x(n− 1) and yn = −d
2 +∆y(n− 1). Substituting (3.44) into (3.31) and
testing with piecewise linear basis functions yields
N∑n=1
Jn
jk0Z0
ac
xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′)e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dy′ dx′ dy dx
− Λn(x)Λn(y)jk0Y0 (εr(x, y)− 1)
= 0. (3.47)
The impedance matrix elements shown in Appendix C can be determined evaluating the
integrals in (3.47).
The transforms of the testing and expansion functions, denoted Λm(kx) and Λn(kx) are
given in (2.60). The transforms of the testing and expansion functions, denoted Λm(ky)
and Λn(ky) are
Λm(ky) = ∆ sinc2(ky∆2
)e−jkyym
Λn(ky) = ∆ sinc2(ky∆2
)ejkyyn .
The transforms of the two “half-basis” functions, denoted Λ1(kx) and ΛN (kx), are given in
(2.61). The transforms of the two “half-basis” functions, denoted Λ1(ky) and ΛN (ky), are
Λ1(ky) = ∆ejkyy1
(ky∆)2(1 + jky∆− ejky∆
)
ΛN (ky) = ∆ejkyyN
(ky∆)2(1− jky∆− e−jky∆
).
TEz Case
Piecewise linear expansion / Piecewise linear testing For piecewise constant func-
tions, we discretize the dielectric region into N=NxNy subsections of width ∆x = b/(Ny−1)
63
and ∆y = d/(Ny − 1), respectively, that approximate the original surface and approximate
J = xJx+ yJy by a linear combination of N piecewise linear basis functions with unknown
current coefficients Jxn , Jyn
Jx =N∑n=1
JxnΛn(x) (3.48a)
Jy =N∑n=1
JynΛn(y) (3.48b)
with Λn(x) and Λn(y) are given in (3.45) and (3.46), respectively. Substituting (3.48) into
(3.36) and testing with piecewise linear basis functions produces
N∑n=1
Jxn
jk0Z0
(1 +
1k20
∂2
∂x2
)
×xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′ +Λn(x)Λn(y)
jk0Y0 (εr(x, y) − 1)
+ Jyn
Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
= Eix(x, y)
(3.49a)
N∑n=1
Jxn
Z0
jk0
∂2
∂x∂y
xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
+ Jyn
jk0Z0
(1 +
1k20
∂2
∂y2
) xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
+Λn(x)Λn(y)
jk0Y0 (εr(x, y)− 1)
= Ei
y(x, y). (3.49b)
The impedance matrix elements shown in Appendix C can be obtained by evaluating the
coupled integrals in (3.49).
Piecewise constant expansion / Piecewise constant testing For piecewise constant
functions, we discretize the dielectric region into N=NxNy subsections of width ∆x = b/Nx
and ∆y = d/Ny, respectively, to approximate the original surface and represent J = xJx +
yJy by a linear combination of N piecewise constant basis functions with unknown current
64
coefficients Jxn , Jyn
Jz(x, y) =N∑n=1
JnΠn(x, y) =N∑n=1
JnΠn(x)Πn(y) (3.50)
with Πn(x) and Πn(y) are given in (3.38) and (3.39), respectively. Substituting (3.50) into
(3.36) and testing with piecewise linear basis functions produces
N∑n=1
Jxn
jk0Z0
(1 +
1k20
∂2
∂x2
)
×xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′ +Λn(x)Λn(y)
jk0Y0 (εr(x, y) − 1)
+ Jyn
Z0
jk0
∂2
∂x∂y
b2∫
− b2
d2∫
− d2
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
= Eix(x, y)
(3.51a)
N∑n=1
Jxn
Z0
jk0
∂2
∂x∂y
xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
+ Jyn
jk0Z0
(1 +
1k20
∂2
∂y2
) xn+∆x∫xn−∆x
yn+∆y∫yn−∆y
Λn(x′)Λn(y′)∑p,q
e−jkxp(x−x′) e−jkyq (y−y′)
k20 − k2
xp− k2
yq
dx′ dy′
+Λn(x)Λn(y)
jk0Y0 (εr(x, y)− 1)
= Ei
y(x, y). (3.51b)
The impedance matrix elements shown in Appendix C can be obtained by evaluating the
coupled integrals in (3.51).
3.3.3 Matrix Solution
In order to accurately determine the eigenvalues of the impedance matrices given in
Appendix C, a sufficient number of Floquet modes Np in the x direction and Nq in the y
direction and subsectional unknowns N=NxNy must be included. The minimization proce-
dure outlined in Chapter 2 is carried out for two-dimensional periodic media by fixing the
frequency of operation and one of the two propagation constants kx0 or ky0 and allowing the
optimization code to vary the other propagation constant until a minimum is found. If more
than one minimum (eigenvalue) is found, implying that multiple modes are propagating in
65
the structure, the corresponding eigenvectors are analyzed to determine which eigenvalue
corresponds to which propagating mode. A nontrivial solution for the fields requires the
matrix determinant to be zero, which results in a characteristic equation. The eigenvalues
(propagation constants) kx0 , ky0 are obtained from the roots of this equation. For a loss-
less structure, the propagation constant of a guided wave is a real number, however, in the
stopbands, the propagation constant is complex-valued.
In order to verify that the solution of the propagation constants computed using the
method of moments technique and PWE solution is correct, the band structure of a periodic
array of dielectric rods is calculated for a filling fraction b/a=d/c=0.3545 and relative di-
electric constant εr=8.9. Figure 3.9 shows the calculated band structure for both the TMz
and TEz cases. The moment method solution includes 225 subsectional bases (Nx=Ny=15)
and 961 Floquet modes (Np=Nq=31). The Fourier series solution uses 121 Floquet modes.
It is clear that the solutions obtained by the two methods are again in agreement with each
other and are also in agreement with [38, 129].
βx (βy=0) βy (βx=π/a) βx=βy
Normalized propagation constant ( βa)
0.00 3.14 6.28 9.42
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
TM (FS)TE (FS)TM (MoM)TE (MoM)
Figure 3.9: Band structure (TMz, TEz) of a periodic lattice of dielectric columns with fillingfraction b/a=d/c=0.3545, εr=8.9, Nx=Ny=15, and Np=Nq=31
Similar conclusions for two-dimensional dielectric structures as those expressed for one-
dimensional structures can be drawn about the number of unknowns needed to accurately
model the dielectric material and the number of requisite Floquet modes to include in the
double series to accurately determine the propagation constants. In Tables 2.1 – 2.4 of
66
Chapter 2, the accuracy of the propagation constant determined for both the TEx and
TMx cases increases both by increasing the number of Floquet mode contributions and
by modeling the equivalent current more carefully by increasing the number of unknowns.
A simple pulse basis expansion serves to solve the problem quickly and accurately. This
observation holds true as well for two-dimensional dielectric structures. As expected, the
accuracy of the propagation constants determined for both the TMz and TEz cases increases
with increasing Floquet contributions and with increasing subsectional bases. Additionally,
the advantage of incorporating higher order basis functions does not significantly increase
the accuracy and actually increases the cost and complication of numerical implementation.
Tables 2.1 – 2.4 in Chapter 2 clearly show that the accuracy of computing the propagation
constant is more sensitive to the number of subsectional bases used in the approximation
of the current than the number of Floquet modes used to compute the Floquet series. The
computational cost of computing each individual element in the impedance matrix in two
dimensions is proportional to NpNq. In two dimensions, the total cost of computing the full
impedance matrix is proportional to N2xN
2yNpNq. Thus, incorporating fewer basis functions
decreases the cost significantly.
As was mentioned previously in Chapter 2, the restriction that the IE/MoM solution
must compute the impedance matrix for each iteration of the optimizer requires significant
computational time. Consequently, for the band structure of an infinite lattice of dielectric
inserts in a background of differing dielectric constant, the solution obtained using the
Fourier series solution is the formulation of choice. The derivation and implementation of
the IE/MoM solution developed in Section 3.3 is extended to three-dimensional structures
in Chapter 4.
3.4 Effective Medium Theory (EMT)
To simplify the computation of the band structure for two-dimensionally periodic me-
dia, effective medium theory is applied to reduce the two-dimensional periodic structure
to a one-dimensional equivalent structure. This useful technique can be applied when the
period of the structure is much smaller than the wavelength. Using EMT, each periodic
row of the two-dimensional structure is replaced by a thin homogeneous layer of effective
permittivity [50] which is determined solely as a function of the geometrical and electrical
parameters of the lattice.
Consider the two-dimensional lattice with dielectric constant ε2 immersed in a back-
67
x
y
a
b
c d
ε eff
ε eff
ε eff
ε eff
ε2
ε1
Figure 3.10: Effective medium theory
ground with dielectric constant ε1 and the equivalent one-dimensional lattice with dielectric
constant εeff, illustrated in cross-section in Figure 3.10. Assume a plane wave is incident
along the x direction (Γ–X direction in the BZ). Effective medium theory allows for each
row of the square lattice to be replaced by an equivalent layer of permittivity εeff. For TMz
polarization, where the electric field parallel to z axis, Rytov [82] expands the effective
permittivity εeff in a power series of the period-to-wavelength ratio α=a/λ and is given by
εeff = ε0 +π2
3[f(1− f)(ε2 − ε1)]2 α2 +O(α4) (3.52)
where ε0 is the average relative permittivity ε0 = ε2f − ε1(1 − f), α = a/λ0, and f is the
filling fraction of the medium defined by the ratio of the width b of the effective homogeneous
layer to the lattice constant a. For TEz polarization, where the electric field parallel to y
axis, the effective permittivity is given by
εeff =1a0
+π2
3
[f(1− f)(ε2 − ε1)
ε2ε1
]2 ε0a3
0
α2 +O(α4) (3.53)
where a0 is the arithmetic average of the inverse relative permittivities a0 = f/ε2−(1−f)/ε1and f is defined above.
Two conclusions can be drawn about the results obtained using EMT. First, for large
period-to-wavelength ratios, the EMT approximation fails to accurately produce the band
structure for either the TMz or TEz case. This is expected since the physical boundary
conditions that exist in the real structure are not approximated well. Secondly, as the ratio
decreases, the validity of the EMT approximation increases. Even for a relatively large
period-to-wavelength ratio of 0.5, the results obtained using EMT are surprisingly good.
68
The band structure of the exact two-dimensional structure determined using the plane
wave expansion method and integral equation models outlined in Sections 3.2 and 3.3 and
the one-dimensional equivalent structure modeled using EMT is illustrated in Figure 3.11.
Thus, the lattice of Figure 3.10 with filling fraction b/a=d/c=0.3545 and dielectric constant
εr=8.9 is modeled by an array of dielectric slabs with the same filling fraction as the two-
dimensional array with effective permittivity εeff. For the TMz modes, the EMT results
compare favorably to [129] for normalized frequencies f0 < 1.0. The results obtained for
the TEz modes compares less favorably, particularly for higher frequencies.
If the filling fraction b/a=d/c is decreased to 0.2 and if all other geometrical parameters
remain the same, the ability of the EMT to effectively model the two-dimensional lattice as
an equivalent one-dimensional array can be seen in Figure 3.12 to improve greatly for the
TEz case. Although an improvement over the TMz case is not seen, the EMT effectively
models the first band gap as accurately as might be needed.
3.5 Conclusions
The solution for the propagation of electromagnetic energy through a two-dimensional
periodic dielectric structure has been determined using the plane wave expansion method
and an integral equation/method of moments solution. The plane wave expansion method
solves the two-dimensional (differential) wave equation by expanding the periodic functions
in a Fourier series. The integral equation solution is derived and implemented using polariza-
tion currents for a two-dimensionally periodic media. For infinite two-dimensional dielectric
structures, modeling the dielectric material with increasing number of unknowns increases
the accuracy of the solution as does increasing the number of Floquet modes. The disadvan-
tage of the IE/MoM solution is the significant increase in computational time required to
compute the eigenvalues (propagation constants) for the two-dimensional structures. The
advantage of the IE/MoM solution is its straightforward extension to three-dimensional
layered periodic structures with periodic material implants.
An effective medium theory (EMT) approximation for the two-dimensional lattice has
been shown to effectively model propagation in specific directions within the lattice struc-
ture. The use of EMT to effectively approximate propagation through a periodic structure
is significant since, for many of the applications used in microwave devices, the first (and
lowest) band structure is the band of interest. The approximation is particularly helpful
when one considers the complication of a two-dimensional formulation. Unfortunately, the
69
Normalized propagation constant (k a)
0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.25
0.50
0.75
1.00
TM modesTM EMT
(a)
Normalized propagation constant (k xa)
0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.25
0.50
0.75
1.00
TE modesTE EMT
(b)
Figure 3.11: Exact band structure and effective medium theory approximation for periodiclattice of dielectric columns with filling fraction b/a=d/c=0.3545, dielectricconstant εr=8.9, and period-to-wavelength ratio α = a/λ0=0.5 for (a) TMz
case and (b) TEz case
EMT is restricted to near normal incidence precluding its use for many applications of
interest.
70
Normalized propagation constant (k xa)
0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.25
0.50
0.75
1.00
TE modesTE EMT
(a)
Normalized propagation constant (k a)
0.00 1.57 3.14
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.00
0.25
0.50
0.75
1.00
TM modesTM EMT
(b)
Figure 3.12: Exact band structure and effective medium theory approximation for periodiclattice of dielectric columns with filling fraction b/a=d/c=0.2, dielectric con-stant εr=8.9, and period-to-wavelength ratio α = a/λ0=0.5 for (a) TMz caseand (b) TEz case
To validate the usefulness and realizability of microwave devices that incorporate sub-
strate materials with periodicities in two directions, the design, fabrication, and implemen-
71
tation of a two-dimensional periodic dielectric structure developed for use in parallel-plate
mode reduction in conductor-backed slots is presented extensively in Appendix D. Band
structures are designed to produce the requisite substrate properties using the PWE method
and IE/MoM solutions developed in this chapter.
72
CHAPTER 4
Scattering From an Inhomogeneous Doubly Periodic
Dielectric Layer Above a Layered Medium
4.1 Introduction
In this chapter, the solution found in [84] for the two-dimensional scattering from an
inhomogeneous periodic layer above a half-space layered medium is extended to general
three-dimensional scattering from an inhomogeneous doubly periodic layer above a half-
space layered medium. The solution of this problem is derived and implemented using a full-
wave integral equation/method of moments approach similar to the formulation for the one-
and two-dimensional periodic structures of Chapters 2 and 3. In this formulation, coupled
integral equations are derived that incorporate both the two-dimensional planar periodic
free-space Green’s function (PFSGF) and the inclusion of three-dimensional material blocks.
Equivalent polarization currents are used to model the material inclusions and the method
of moments is used to discretize the geometry and the integral equations. The resulting
matrix equation is numerically solved for various quantities of interest. First, the solution
of plane wave scattering from an inhomogeneous doubly periodic dielectric layer over a
layered medium is derived and implemented in Section 4.2. The reflection coefficient is
determined for arbitrary polarization (vertical, horizontal, or a combination thereof) and
arbitrary direction of incidence (φ0 ∈ [0, 2π], θ0 ∈ [0, π]) for a number of sample structures
and validated with results obtained from canonical problems and results published in the
open literature.
Of particular concern in the solution of the periodic structure is the convergence of
the resulting Floquet series. Although the contribution of the off-plane periodic elements
converges quickly, the convergence of the on-plane periodic elements is notoriously slow. In
73
order to compute the impedance matrix elements in a reasonable amount of time, various
series acceleration techniques and transformations have been suggested including the Pois-
son transformation [70], Kummer’s method [51], the Shanks’ transform [90], and the Ewald
transformation [30]. Summation acceleration techniques convert a slowly converging series
to a rapidly converging one by allowing the series to be transformed into a second series that
converges to the same limit but does so in a rapid fashion. The aforementioned transforms
have all been implemented in conjunction with the slowly-converging PFSGF and each
transform has its advantages and disadvantages when applied to the PFSGF in a particular
fashion. However, some of the transforms have only been implemented for limited cases
of the PFSGF. For the problems of interest in this chapter, a Poisson transformation and
a Shanks’ transformation are successfully implemented to improve both the speed and the
accuracy of the impedance matrix element computations. Additionally, Kummer’s method
is applied to specific series to compare the convergence rate of these different series accel-
eration techniques. A complete treatment of these and other series acceleration techniques
is found in Section 4.5.
4.2 Formulation of Integral Equations
In this section, volume integral equations (IE) are derived from Maxwell’s equations
and the boundary conditions that when solved yield equivalent currents from which total
scattered fields can be determined. The integral equations are discretized and cast into a
matrix equation form through the use of the method of moments (MoM) numerical solution
technique. The unknowns are obtained from the solution of the resulting linear system.
4.2.1 Derivation of Electric Field Integral Equation
The derivation of the electric field integral equation outlined in Chapter 2 is repeated
below for convenience. The total electric field is viewed as the sum of an incident field
Ei(r) due to radiation from a known source with the scatterer absent and a scattered field
E [J; r;V ] which is due to radiation by equivalent volume currents J which reside in a volume
V
E(r) = E [J; r;V ] + Ei(r), (4.1)
in which the operator E [J; r;V ] can be expressed in terms of a Hertz potential [108] as
E [J; r;V ] = k20 Π(r) +∇∇ ·Π(r) (4.2a)
74
where
Π(r) = − Z0
8π2k0
∫∫∫V
J(r′)∫∫∞
e−jkz|z−z′|
kze−jky(y−y′)e−jkx(x−x′) dkx dky dr′ (4.2b)
with
kz =
√k20 − k2
x − k2y k2
0 > k2x + k2
y,
−j√k2x + k2
y − k20 k2
0 < k2x + k2
y.(4.2c)
and Z0=1/Y0 is the intrinsic impedance of free-space. The electric field must satisfy
E [J; r;V ] + Ei(r) = E(r), r ∈ V. (4.3)
4.2.2 Integral Equations
Plane wave excitation
A representative grounded doubly periodic dielectric layer with period a in the x direc-
tion and period c in the y direction is illustrated in Figure 4.1. The perforated dielectric
a
cx
y
z
Figure 4.1: Doubly periodic dielectric layer with period a in the x direction and period cin the y direction over a layered medium
layer is excited by an incident time harmonic (ejωt) plane wave defined by
Ei = Pi e−jki·r =
(ehhi + evvi
)e−jki·r (4.4)
where Pi = ehhi+evvi denotes the polarization of the incident wave and ki is the direction
of propagation. The unit vectors hi, vi, and ki are defined in terms of the spherical angles
75
(φ0, θ0) as
ki = x cosφ0 sin θ0 + y sinφ0 sin θ0 − z cos θ0 (4.5a)
hi =ki × z
|ki × z| = x sinφ0 − y cosφ0 (4.5b)
vi = hi × ki = x cosφ0 cos θ0 + y sinφ0 cos θ0 + z sin θ0 (4.5c)
and are shown in Figure 4.2. The kDB coordinate system [45] consists of the wave vector
k and the plane containing the electric and magnetic flux density vectors D and B. In
z
y
x
φ
θ
hk
v
i
i
i
h s
vs ks
φ
θo
o
s
s
Figure 4.2: Coordinate system (kDB) for incident and scattered field
the absence of the dielectric layer (to be replaced by equivalent polarization currents), the
reflected field from a surface can be easily computed as
Er = Pre−jks·r =
(Rhehhs +Rvevvs
)e−jks·r (4.6)
where Rh and Rv are the Fresnel reflection coefficients for horizontal and vertical polarized
electric fields, respectively, and ks is the direction of propagation. The unit vectors hs, vs,
and ks are defined in terms of the spherical angles (φs, θs) as
ks = x cosφs sin θs + y sinφs sin θs + z cos θs (4.7a)
hs =ks × z
|ki × z| = x sinφs − y cosφs (4.7b)
vs = hs × ks = −x cosφs cos θs − y sinφs cos θs + z sin θs (4.7c)
76
Note that ks = ki − 2(z · ki)z.The total electric field everywhere is computed as the sum of the incident and reflected
electric fields in (4.4) and (4.6) and the scattered field Es(r) produced by the equivalent
induced electric current. Thus, (4.3) becomes upon substitution
J(r)jk0Y0 (εr(r)− 1)
= Ei(r) + Er(r) + Es(r)
=(ehhi + evvi
)e−jki·r +
(Rhehhs +Rvevvs
)e−jks·r + Es(x, y, z). (4.8)
The three coupled integral equations from (4.8) become
Jx(r)jk0Y0 (εr(r)− 1)
=(ehx · hi + evx · vi
)e−jki·r +
(Rhehx · hs +Rvevx · vs
)e−jks·r
+∫∫∫∞
Gxx(r; r′)Jx(r′) +Gxy(r; r′)Jy(r′) +Gxz(r; r′)Jz(r′)
dr′ (4.9)
Jy(r)jk0Y0 (εr(r)− 1)
=(ehy · hi + evy · vi
)e−jki·r +
(Rhehy · hs +Rvevy · vs
)e−jks·r
+∫∫∫∞
Gyx(r; r′)Jx(r′) +Gyy(r; r′)Jy(r′) +Gyz(r; r′)Jz(r′)
dr′ (4.10)
Jz(r)jk0Y0 (εr(r)− 1)
= evz · vie−jki·r +Rvevz · vse−jks·r
+∫∫∫∞
Gzx(r; r′)Jx(r′) +Gzy(r; r′)Jy(r′) +Gzz(r; r′)Jz(r′)
dr′ (4.11)
where Gxx—Gzz are the components of the dyadic free-space Green’s function.
Since the dielectric material is periodic in x with period a and prescribed phase shift
kx0a and periodic in y with period c and prescribed phase shift ky0c, the resulting equivalent
currents must satisfy
J(x+ pa, y + qc, z) = J(x, y, z)e−jkx0pae−jky0qc (4.12)
where
kx0 = k0 cosφ0 sin θ0
ky0 = k0 sinφ0 sin θ0
77
and p,q are the Floquet indices in the x - and y-directions, respectively. The x -component
of the scattered field integral, Ixx, in (4.9) is defined by
Ixx =∫∫∫∞
Gxx(r; r′)Jx(r′) dr′
=∑p,q
(p+ 12)a∫
(p− 12)a
(q+ 12)c∫
(q− 12)c
∞∫−∞
Gxx(r; r′)Jx(r′)e−jkx0pae−jky0qc dr′
=
a2∫
− a2
c2∫
− c2
∞∫−∞
Gpxx(r; r′)Jx(r′) dr′ (4.13)
where the periodic free-space Green’s function, denoted Gp(r, r′), is defined in terms of the
free-space Green’s function as
Gp(r; r′) =∑p,q
G(x, y, z;x′ + pa, y′ + qc, z′)e−jkx0pae−jky0qc. (4.14)
The significance of this procedure is that the solution of the periodic structure has now been
reduced to solving that of the single unit cell shown in Figure 4.3. Explicitly, Gpxx(r; r′) in
b
d
a
c
Figure 4.3: Doubly periodic dielectric layered unit cell
(4.13) can be written as
Gpxx(r; r′) = − Z0
8π2k0
∑p,q
(k20 +
∂2
∂x2
)∫∫∞
[e−jkz|z−z′|
kz+Rx
e−jkz(z+z′)
kz
]
× e−jky(y−y′−qc)e−jkx(x−x′−pa)e−jkx0pae−jky0qc dkx dky, (4.15)
78
where Rx = Rh. If the above is rewritten as
Gpxx(r; r′) = − Z0
8π2k0
∑p,q
(k20 +
∂2
∂x2
)∫∫∞
[e−jkz|z−z′|
kz+Rx
e−jkz(z+z′)
kz
]
× e−jkx(x−x′)e−jky(y−y′)e−j(kx0−kx)pae−j(ky0−ky)qc dkx dky, (4.16)
an important observation can be made which greatly simplifies the formulation. From the
theory of Fourier Series we can represent a periodic train of Dirac delta functions by an
infinite summation of complex exponentials,∑p
e−j(kx0−kx)pa =2πa
∑p
δ
[kx − kx0 −
2πpa
](4.17a)
∑q
e−j(ky0−ky)qc =2πc
∑q
δ
[ky − ky0 −
2πqc
]. (4.17b)
Substituting the Dirac delta functions representation in (4.17) and carrying out the inte-
grations in (4.16) yields
Gpxx(r; r′)
= − Z0
2k0ac
∑p,q
(k20 +
∂2
∂x2
)[e−jkzpq |z−z′|
kzpq
+Rxe−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18a)
where
kxp = kx0 +2πpa
= k0 cosφ0 sin θ0 +2πpa
kyq = ky0 +2πqc
= k0 sinφ0 sin θ0 +2πqc
with
kzpq =
√k20 − k2
xp− k2
yq, k2
0 > k2xp
+ k2yq,
−j√k2xp
+ k2yq
− k20 , k2
0 < k2xp
+ k2yq.
The remaining periodic free-space Green’s function components Gpxy–Gpzz are defined by
Gpxy(r; r′) = − Z0
2k0ac
∑p,q
∂2
∂x∂y
[e−jkzpq |z−z′|
kzpq
+Rye−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18b)
Gpxz(r; r′) = − Z0
2k0ac
∑p,q
∂2
∂x∂z
[e−jkzpq |z−z′|
kzpq
+Rze−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18c)
79
Gpyx(r; r′) = − Z0
2k0ac
∑p,q
∂2
∂y∂x
[e−jkzpq |z−z′|
kzpq
+Rxe−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18d)
Gpyy(r; r′)
= − Z0
2k0ac
∑p,q
(k20 +
∂2
∂y2
)[e−jkzpq |z−z′|
kzpq
+Rye−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18e)
Gpyz(r; r′) = − Z0
2k0ac
∑p,q
∂2
∂y∂z
[e−jkzpq |z−z′|
kzpq
+Rze−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18f)
Gpzx(r; r′) = − Z0
2k0ac
∑p,q
∂2
∂z∂x
[e−jkzpq |z−z′|
kzpq
+Rxe−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18g)
Gpzy(r; r′) = − Z0
2k0ac
∑p,q
∂2
∂z∂y
[e−jkzpq |z−z′|
kzpq
+Rye−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18h)
Gpzz(r; r′)
= − Z0
2k0ac
∑p,q
(k20 +
∂2
∂z2
)[e−jkzpq |z−z′|
kzpq
+Rze−jkzpq (z+z′)
kzpq
]e−jkxp(x−x′)e−jkyq (y−y′)
(4.18i)
where Ry = Rh and Rz = Rv. If the equivalent currents are assumed to radiate in the
presence of a ground plane, the Green’s functions listed in (4.18a)–(4.18i) can be easily
manipulated to agree with similar expressions in [133] by setting Rh=−1 and Rv=+1.
Reflection from a Dielectric Layer
If the ground plane backing the doubly periodic dielectric layer is replaced by a layered
medium, perhaps also grounded, the Fresnel reflection coefficients (Rh and Rv) are no
longer simply ±1. The geometry of the layered medium is shown in Figure 4.4 where the
80
region 1
region 2
region N
region 0
region (N+1)
θο
z=d
d
d
d
d
0
1
2
(N-1)
N
ε
ε
ε
ε
ε
0
1
2
N
(N+1)
E(H)
Figure 4.4: Plane wave reflection from layered medium
lth interface is the defined by the plane located at z=dl. Region 0 is the region above the
layered medium and region (N + 1) is the semi-infinite region below the N layers. The
solution for the total reflection coefficient of the medium can be obtained by applying the
appropriate boundary conditions to the tangential components of the electric and magnetic
field at each interface. Consequently, the (N + 1) interfaces yield (2N + 2) equations to
solve for the (2N + 2) unknown amplitude coefficients. Explicitly, the (2N + 2) unknowns
are the two unknowns in each of the N layers (one each for the waves traveling in the ±zdirections), one for the unknown transmission coefficient of the semi-infinite region, and one
for the reflection coefficient of interest.
Horizontal polarization Assume the electric field in the lth layer of a medium is polar-
ized in the horizontal direction. From (4.4) and following [84], the electric field in the lth
layer is written in the form
Ehl=[cile
jkzlz + crl e
−jkzlz]e−jkx0xe−jky0y (4.19)
where
kzl= k0
√εl − sin2 θ0 (4.20)
and cil and crl are the amplitudes of the −z and +z traveling waves in the lth layer, respec-
tively, and εl is the permittivity of the lth layer. The magnetic field in the lth layer can
be found by applying (2.1b) to the electric field above. Carrying out the required vector
operations, one finds that the tangential components of the vertically polarized magnetic
81
field in the lth layer have the form
Hvl=
kzl
k0Z0
[cile
jkzlz − crl e−jkzl
z]e−jkx0xe−jky0y. (4.21)
The following recursive relationship that relates the amplitudes of the lth layer to those of
the (l + 1)th layer can be derived by requiring the tangential electric and magnetic field to
be continuous across each of the dielectric interfaces
crlcil
=
(crl+1/c
il+1
)+ Γhl e
−j2kz(l+1)dl(
crl+1/cil+1
)Γhl + e
−j2kz(l+1)dle−j2kzl
dl (4.22)
where
Γhl =kzl
− kz(l+1)
kzl+ kz(l+1)
. (4.23)
The solution for the total horizontal reflection coefficient Rh is initiated by assuming
ci0=1 in region 0 and by noting that crN+1=0 in the semi-infinite region. Starting from
cr(N+1)/ci(N+1)=0, equations (4.22) and (4.23) are used repeatedly to solve for the amplitudes
of the waves in each layer beginning with the interface located at z = dN and continuing to
the interface between the zeroth and first layer. The horizontal reflection coefficient is then
determined as Rh=cr0/ci0.
Vertical polarization If the electric field is vertically polarized, the same recursive re-
lation in (4.22) is found but with Γhl replaced by
Γvl =ε(l+1)kzl
− εlkz(l+1)
ε(l+1)kzl+ εlkz(l+1)
. (4.24)
The solution for the vertical reflection coefficient is found from Rv=cr0/ci0.
Grounded medium If the N th layer in the medium is grounded, ΓhN=−1 and ΓvN=+1,
and the process is executed as before.
4.3 Numerical Implementation
The first step in the method of moments (MoM) numerical method [34] is to discretize
the geometry and approximate the unknown electric current in the dielectric region with
either subsectional or entire domain basis functions. The equivalent polarization currents
are expanded in a linear combination of these basis functions, and the equations are tested
82
in order to obtain an adequate number of equations to solve for the unknown coefficients of
the basis functions.
The dielectric region is discretized into N=NxNyNz piecewise constant subvolumes of
width ∆x = b/Nx, length ∆y = d/Ny, and depth ∆z = t/Nz. The current J = xJx +
yJy + zJz is expanded in a linear combination of N piecewise constant basis functions
with unknown current coefficients Jxn , Jyn , Jzn centered at xn = − b2 + ∆x(n − 1
2), yn =
−d2 +∆y(n− 1
2), and zn = ∆z(n− 12)
Jx =N∑n=1
JxnΠn(x, y, z) =N∑n=1
JxnΠn(x)Πn(y)Πn(z) (4.25a)
Jy =N∑n=1
JynΠn(x, y, z) =N∑n=1
JynΠn(x)Πn(y)Πn(z) (4.25b)
Jz =N∑n=1
JznΠn(x, y, z) =N∑n=1
JznΠn(x)Πn(y)Πn(z) (4.25c)
where
Πn(x) =
1, xn − ∆x
2 < x < xn + ∆x2 ;
0, otherwise;(4.26)
and
Πn(y) =
1, yn − ∆y
2 < y < yn +∆y
2 ;
0, otherwise;(4.27)
and
Πn(z) =
1, zn − ∆z
2 < z < zn + ∆z2 ;
0, otherwise.(4.28)
4.3.1 Impedance Matrices
The scattered field is determined by integrating the equivalent currents over the periodic
Green’s function. By discretizing the resulting coupled equations, they can be cast into the
following matrix form Zxxmn Zxymn Zxzmn
Zyxmn Zyymn Zyzmn
Zzxmn Zzymn Zzzmn
Jxn
Jyn
Jzn
=
V xm
V ym
V zm
. (4.29)
83
Specifically, for the Zxx submatrix, the current representation is inserted into (4.9) and
the integrations are carried out. Testing the resulting equation using point collocation, we
obtain for all off-plane, i.e. zm = zn, elements
Zxx =
a2∫
− a2
c2∫
− c2
t∫0
Gpxx(r; r′) Jx(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
(k20 +
∂2
∂x2
)[e−jkzpq |z−z′|
kzpq
+Rxe−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jx(x′, y′, z′) dx′ dy′ dz′ (4.30)
Zxxmn = −Z0∆x∆y∆z
2k0ac
∑p,q
(k20 − k2
xp
kzpq
)[e−jkzpq |zm−zn| +Rxe−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.31)
For zm = zn, the integration over z must be computed carefully. The resulting on-plane
impedance matrix elements for Zxx are
Zxxmn = −Z0∆x∆y∆z
2k0ac
∑p,q
(k20 − k2
xp
kzpq
)
×[(2− 2e−jkzpq∆z/2
)kzpq∆z
+Rx sinc(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn). (4.32)
The remaining impedance matrix elements Zxymn–Zzzmn are explicitly below given for on-plane
and off-plane elements.
Submatrix Zxy
Zxy =
a2∫
− a2
c2∫
− c2
t∫0
Gpxy(r; r′) Jx(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
∂2
∂x∂y
[e−jkzpq |z−z′|
kzpq
+Rye−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jx(x′, y′, z′) dx′ dy′ dz′ (4.33)
84
For zm = zn,
Zxymn =Z0∆x∆y∆z
2k0ac
∑p,q
kyqkxp
kzpq
[e−jkzpq |zm−zn| +Rye−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.34)
For zm = zn,
Zxymn =Z0∆x∆y∆z
2k0ac
∑p,q
kyqkxp
kzpq
[(2− 2e−jkzpq∆z/2)
kzpq∆z+Ry sinc
(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.35)
Submatrix Zyx
Zyx =
a2∫
− a2
c2∫
− c2
t∫0
Gpyx(r; r′) Jy(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
∂2
∂y∂x
[e−jkzpq |z−z′|
kzpq
+Rxe−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jy(x′, y′, z′) dx′ dy′ dz′ (4.36)
For zm = zn,
Zyxmn =Z0∆x∆y∆z
2k0ac
∑p,q
kyqkxp
kzpq
[e−jkzpq |zm−zn| +Rxe−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.37)
For zm = zn,
Zyxmn =Z0∆x∆y∆z
2k0ac
∑p,q
kyqkxp
kzpq
[(2− 2e−jkzpq∆z/2)
kzpq∆z+Rx sinc
(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.38)
85
Submatrix Zyy
Zyy =
a2∫
− a2
c2∫
− c2
t∫0
Gpyy(r; r′) Jy(x′, y′, z′) dx′ dy′ dz′
= −k0Z0
2ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
(1 +
1k20
∂2
∂y2
)[e−jkzpq |z−z′|
kzpq
+Rye−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jy(x′, y′, z′) dx′ dy′ dz′ (4.39)
For zm = zn,
Zyymn = −Z0∆x∆y∆z
2k0ac
∑p,q
(k20 − k2
yq
kzpq
)[e−jkzpq |zm−zn| +Rye−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.40)
For zm = zn,
Zyymn = −Z0∆x∆y∆z
2k0ac
∑p,q
(k20 − k2
yq
kzpq
)
×[(2− 2e−jkzpq∆z/2
)kzpq∆z
+Ry sinc(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.41)
Submatrix Zxz
Zxz =
a2∫
− a2
c2∫
− c2
t∫0
Gpxz(r; r′) Jx(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
∂2
∂x∂z
[e−jkzpq |z−z′|
kzpq
+Rze−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jx(x′, y′, z′) dx′ dy′ dz′ (4.42)
86
For zm = zn,
Zxzmn =Z0∆x∆y∆z
2k0ac
∑p,q
kxp
[sgn (zn − zm)e−jkzpq |zm−zn| +Rze−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.43)
For zm = zn,
Zxzmn =Z0∆x∆y∆z
2k0ac
∑p,q
kxp
[Rz sinc
(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.44)
Submatrix Zzx
Zzx =
a2∫
− a2
c2∫
− c2
t∫0
Gpzx(r; r′) Jz(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
∂2
∂z∂x
[e−jkzpq |z−z′|
kzpq
+Rxe−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jz(x′, y′, z′) dx′ dy′ dz′ (4.45)
For zm = zn,
Zzxmn =Z0∆x∆y∆z
2k0ac
∑p,q
kxp
[sgn (zn − zm)e−jkzpq |zm−zn| +Rxe−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.46)
For zm = zn,
Zzxmn =Z0∆x∆y∆z
2k0ac
∑p,q
kxp
[Rx sinc
(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.47)
87
Submatrix Zyz
Zyz =
a2∫
− a2
c2∫
− c2
t∫0
Gpyz(r; r′) Jy(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
∂2
∂y∂z
[e−jkzpq |z−z′|
kzpq
+Rze−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jx(x′, y′, z′) dx′ dy′ dz′ (4.48)
For zm = zn,
Zyzmn =Z0∆x∆y∆z
2k0ac
∑p,q
kyq
[sgn (zn − zm)e−jkzpq |zm−zn| +Rze−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.49)
For zm = zn,
Zyzmn =Z0∆x∆y∆z
2k0ac
∑p,q
kyq
[Rz sinc
(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.50)
Submatrix Zzy
Zzy =
a2∫
− a2
c2∫
− c2
t∫0
Gpzy(r; r′) Jz(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
∂2
∂z∂y
[e−jkzpq |z−z′|
kzpq
+Rye−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jz(x′, y′, z′) dx′ dy′ dz′ (4.51)
For zm = zn,
Zzymn =Z0∆x∆y∆z
2k0ac
∑p,q
kyq
[sgn (zn − zm)e−jkzpq |zm−zn| +Rye−jkzpq (zm+zn)
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.52)
88
For zm = zn,
Zzymn =Z0∆x∆y∆z
2k0ac
∑p,q
kyq
[Ry sinc
(kzpq∆z
2
)e−jkzpq2zm
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.53)
Submatrix Zzz
Zzz =
a2∫
− a2
c2∫
− c2
t∫0
Gpzz(r; r′) Jz(x′, y′, z′) dx′ dy′ dz′
= − Z0
2k0ac
a2∫
− a2
c2∫
− c2
t∫0
∑p,q
(k20 +
∂2
∂z2
)[e−jkzpq |z−z′|
kzpq
+Rze−jkzpq (z+z′)
kzpq
]
× e−jkxp(x−x′)e−jkyq (y−y′) Jz(x′, y′, z′) dx′ dy′ dz′ (4.54)
For zm = zn,
Zzzmn = −Z0∆x∆y
2k0ac
∑p,q
sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn)
×[ zn+∆z
2∫zn−∆z
2
(k20 − ∂2
∂z∂z′
)e−jkzpq |zm−z′|
kzpq
dz′ +
zn+∆z2∫
zn−∆z2
(k20 +
∂2
∂z2
)Rze−jkzpq (zm+z′)
kzpq
dz′]
(4.55)
Carrying out the integrations in (4.55) leads to the following off-plane elements of the Zzzmnsubmatrix
Zzzmn = −Z0∆x∆y∆z
2k0ac
∑p,q
(k20 − k2
zpq
)[e−jkzpq |zm−zn|
kzpq
+Rze−jkzpq (zm+zn)
kzpq
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)sinc
(kzpq∆z
2
)e−jkxp(xm−xn)e−jkyq (ym−yn). (4.56)
For zm = zn, particular care must be taken to evaluate (4.54). If in the computation the on-
plane matrix elements for the Zzz submatrix the second derivative in (4.54) is not evaluated
carefully, erroneous results will be produced [76]. From distribution theory, carrying out
the inner derivative of the first integral in (4.55) yields
∂
∂z′e−jkzpq |z−z′| = jkzpq sgn(z − z′)e−jkzpq |z−z′|. (4.57)
89
where
sgn(z − z′) =
+1, z > z′;
−1, z < z′.(4.58)
Applying the outer derivative on the result obtained in (4.57) gives
∂
∂z
sgn(z − z′)e−jkzpq |z−z′|
=[−jkzpq + 2δ(z − z′)] e−jkzpq |z−z′|. (4.59)
Consequently,
∂2
∂z∂z′e−jkzpq |z−z′| = jkzpq
[−jkzpq
+ 2δ(z − z′)]e−jkzpq |z−z′|
=[k2zpq
+ 2jkzpqδ(z − z′)]e−jkzpq |z−z′|. (4.60)
Inserting the above into (4.54) and evaluating the first integral produces
Zzzmn = −Z0∆x∆y
2k0ac
∑p,q
sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn)
×[Rz
(k20 − k2
zpq
) e−j2kzpqzm
kzpq
∆z sinc(kzpq∆z
2
)+ k2
0
zn+∆z2∫
zn−∆z2
e−jkzpq |z−z′|
kzpq
dz′
−zn+∆z
2∫zn−∆z
2
[k2zpq
+ 2jkzpqδ(z − z′)] e−jkzpq |z−z′|
kzpq
dz′]. (4.61)
Rewriting the above as
Zzzmn = −Z0∆x∆y
2k0ac
∑p,q
sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn)
×[Rz
(k20 − k2
zpq
) e−j2kzpqzm
kzpq
∆z sinc(kzpq∆z
2
)+(k20 − k2
zpq
) (2− 2e−jkzpq∆z/2)
k2zpq
−zn+∆z
2∫zn−∆z
2
2jkzpqδ(z − z′)e−jkzpq |z−z′|
kzpq
dz′]
(4.62)
and carrying out the remaining integration yields the on-plane matrix elements for the Zzz
submatrix
Zzzmn = −Z0∆x∆y
2k0ac
∑p,q
sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn)
×[Rz
(k20 − k2
zpq
) e−j2kzpqzm
kzpq
∆z sinc(kzpq∆z
2
)+(k20 − k2
zpq
) (2− 2e−jkzpq∆z/2)
k2zpq
−2j]
(4.63)
90
A similar expression to (4.63) can be found by using dyadic analysis [45] and performing
some algebraic manipulation.
4.3.2 Excitation Matrices
The excitation matrix can be found by testing the incidence excitation using point
collocation. For plane wave excitation,
V xm = −
(ehx · hi + evx · vi
)e−jki·rm −
(Rhehx · hs +Rvevx · vs
)e−jks·rm (4.64a)
V ym = −
(ehy · hi + evy · vi
)e−jki·rm −
(Rhehy · hs +Rvevy · vs
)e−jks·rm (4.64b)
V zm = −evz · vie−jki·rm −Rvevz · vse−jks·rm (4.64c)
4.4 Results
A number of simple checks can be performed that illustrate the validity of the formula-
tion and the correctness of the implementation. A grounded dielectric layer can be modeled
using the method of moments solution by setting the filling fractions b/a and d/c to one
(1) and setting the horizontal and vertical reflection coefficients for the ground plane to −1
and +1, respectively, independent of the incidence angle. The moment method solution for
the plane wave scattering from the grounded periodic dielectric structure yields identical
results with the analytical solution obtained for the plane wave scattering of a grounded
layer. The exact horizontal and vertical reflection coefficients at the surface of the grounded
dielectric layer of thickness t and relative permittivity εr are
Rh =kz0 sin kz1t+ jkz1 cos kz1tkz0 sin kz1t− jkz1 cos kz1t
ej2kz0t (4.65a)
Rv =εrkz0 cos kz1t− jkz1 sin kz1tεrkz0 cos kz1t+ jkz1 sin kz1t
ej2kz0 t (4.65b)
where kz0=k0 cos θ0 and kz1=k0√εr − sin2 θ0. Figure 4.5 shows the calculated phase angle
of the reflection coefficient as a function of incidence angle θ0 using the method of moments
solution for 81 subsectional unknowns (Nx=Ny=Nz=3 for each of the three polarizations),
arbitrary plane of incidence φ0=45, unit cell size a=0.1λ0, relative permittivity εr=2.56,
and dielectric thickness t=0.15λd where λd=λ0/√εr. The solutions were determined using
at most 3721 Floquet modes (Np=Nq=61). Also included in the figure is the exact analytical
solution for the two polarizations. The reflection coefficients are evaluated at a height
z0=20t above the ground plane and theoretically should have a magnitude of one (1) for
91
θο (deg)
0.00 30.00 60.00 90.00
Phas
e (d
eg)
-180
-90
0
90
180
RhRvRh (MoM)Rv (MoM)
Figure 4.5: Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=2.56 and dielectric thickness t=0.15λd as a function of θ0 forNx=Ny=Nz=3, Np=Nq=61, and a=0.1λ0
the grounded case. For the data shown in Figure 4.5 where the dielectric is lossless, the
magnitudes of Rh and Rv are 1.00±0.08. The accuracy of the phase values obtained for
Rv degrades slightly at grazing angles near θ0=90. As the angle of incidence increases,
the z -variation in the dielectric cannot be modeled accurately using only 3 unknowns. If
the z -variation is modeled using five (5) or seven (7) unknowns, the error in the magnitude
decreases to less than 2% for all angles of incidence.
Additionally, for the grounded dielectric layer, the value of the reflection coefficients
should be independent of both the plane of incidence and the unit cell size. Identical phases
for the reflection coefficients are calculated for a number of different values of φ0. However,
when the unit cell size is increased, additional unknowns in x - and y-directions must be
used to model the material blocks in order to obtain comparable accuracies to cases where
the unit cell size is smaller. It is interesting to note that for the grounded dielectric layer,
the scattered field produced by the equivalent currents for the two polarizations is exactly
180 out-of-phase. Although the magnitudes and phases of the currents induced by the two
polarizations are quite different, the scattered electric field produced by them is shifted in
phase by π (Rh=−Rv for the ground plane).
As the filling fraction varies from one (1) to zero (0) and unit cell spacings less than a
wavelength, the phase angle of the horizontal and vertical reflection coefficients determined
92
using the method of moments solution should vary between the exact value obtained from
a grounded dielectric layer and the exact value obtained from a ground plane alone. This
is shown in particular in Table 4.1 for a grounded dielectric layer of relative permittivity
εr=2.56, dielectric thickness t=0.15λd, φ0=45, θ0=45, and a=0.25λ0 at z0=20t using 192
subsectional unknowns (Nx=Ny=Nz=4 for each of the three polarizations) and Np=Nq=61.
The exact values for b/a=1 and b/a=0 shown in bold in the table are calculated using the
analytical solution of (4.65).
Table 4.1: Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=2.56 and dielectric thickness t=0.15λd as a function of fillingfraction b/a for Nx=Ny=Nz=4, Np=Nq=61, φ0=45, θ0=45, and a=0.25λ0
b/a Rh Rv
1.00 -64.99 89.52
0.90 -63.42 95.15
0.75 -59.90 103.77
0.50 -56.56 115.48
0.25 -55.01 123.33
0.00 -54.59 125.41
The grounded dielectric slab can also be modeled as a grounded layered medium as
shown in Figure 4.4. This is accomplished by replacing the homogeneous layer of thick-
ness t and relative dielectric constant εr with two layers each of thickness t/2. Region 2
is filled with a homogeneous layer of permittivity εr, region 1 is replaced with equivalent
polarization currents as before, and the ground is located at z=d2. The significant differ-
ence in computation for the layered medium case is the calculation of Rh and Rv. Unlike
the previous example, where the dielectric is replaced by equivalent polarization currents
and the scattering occurs from the ground plane, the reflection coefficients for equivalent
currents radiating over a layered dielectric are, in general, functions of the complex an-
gle, γpq=arctan(kρ/kzpq) where kρ is the phase constant in the transverse plane equal to√k2xp
+ k2yq. Consequently, Rh and Rv must be calculated for every combination of Floquet
indices p, q.The phase angle of the reflection coefficient for the grounded two-layer medium is de-
termined in Figure 4.6 for the same parameters as Figure 4.5. The noticeable increase in
the accuracy of the phase obtained in this solution is attributable to the increased accuracy
93
θο (deg)
0.00 30.00 60.00 90.00
Phas
e (d
eg)
-180
-90
0
90
180
RhRvRh (MoM)Rv (MoM)
Figure 4.6: Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=2.56 and dielectric thickness t=0.15λd as a function of θ0 forNx=Ny=Nz=3, Np=Nq=61, and a=0.1λ0 calculated using the multilayeredsolution
in modeling the dielectric region. The same number of unknowns are used to model the
dielectric layer of thickness t/2 as were used to model the dielectric layer of thickness t in
Figure 4.5. For all angles of incidence, the magnitudes of Rh and Rv are 1.00±0.02. If
the polarization currents in the previous example are arbitrarily set to zero (J=0) or if
the relative permittivity is allowed to approach one (εr → 1) and the solution for the total
scattered field is recomputed, the reflection coefficients should be equivalent to those of a
grounded layer of thickness t/2. This is found to be true in particular for θ0=45 where the
phase of Rh and Rv for both the exact and calculated solution are −55.72 and +110.01,
respectively.
4.5 Series Acceleration Techniques for the Periodic, Free-
Space Green’s Function
Unfortunately, the series listed throughout Section 4.3 have serious but well-documented
convergence issues when z is very close or equal to z′ that must be addressed before an accu-
rate and rapid solution can be found. Summation acceleration techniques convert a slowly
converging series to a rapidly converging one by allowing the series to be transformed into
94
a second series that converges to the same limit but does so in a rapid fashion. A number
of series acceleration techniques have been applied to the periodic, free-space Green’s func-
tion (PFSGF) to accelerate the notorious on-plane convergence problem [42]. Some of the
more useful ones are listed in the Table 4.2. The transformations listed in the first part of
the table are ones implemented in this work; those listed in the second part of the table
are ones that are implemented elsewhere. Each acceleration technique has advantages and
disadvantages and must be applied carefully to the series of interest in order to obtain the
desired results of improved accuracy and increased speed.
Table 4.2: Summary of series transformations
Transformations
Poisson transformation
Shanks’ transformation
Wynn’s ε-algorithm
Kummer transformation
Ewald transformation
ρ-algorithm
θ-algorithm
Chebyshev-Toeplitz (CT) algorithm
Levin’s t-transform
4.5.1 Acceleration Techniques Applied in This Work
Poisson Transformation [70, 51, 99]
The Poisson summation converts a slowly converging series to a rapidly converging one
by allowing the series to be summed in the Fourier transform domain. The reciprocal
spreading property of the Fourier transform says that narrower support in one domain
would necessarily require wider support in the other domain. The transformation can be
written as∞∑
n=−∞f(t+ nT ) =
1T
∞∑n=−∞
e−jnω0tF (nω0) ω0 =2πT. (4.66)
The Poisson transformation transforms the PFSGF into its spectral form which makes it
amenable for use in spectral domain analysis (SDA) of layered media or periodically loaded
95
dielectrics.
To obtain a summation for a two-dimensional array of point current sources, the Poisson
summation formula is first applied to the y coordinate of the three-dimensional Green’s
function
∑p,q
f(p, q) =∑p,q
[(x− pa)2 + (y − qc)2 + z2]1/2 e−jk[(x−pa)2+(y−qc)2+z2]1/2
=∑p,q
12πc
K0
[(2πq
c
)2
− k20
]1/2 [(x− pa)2 + z2]1/2
e−j 2πq
cy. (4.67)
An expression equivalent to a two-dimensional Green’s function can be recovered by ma-
nipulation of the above expression giving
∑p,q
f(p, q) =∑p,q
14cjH
(2)0
([k2 −
(2πqc
)]1/2 [(x− pa)2 + z2]1/2
)e−j
2πqcy. (4.68)
Applying the Poisson summation formula again, but this time to the x coordinate gives the
following Poisson summation formula for the Green’s function f(p, q):
∑p,q
f(p, q) =12ac
∑p,q
e−|z|
[( 2πp
a )2+( 2πqc )2−k2
]1/2
√(2πpa
)+(
2πqc
)− k2
e−j2πp
axe−j
2πqcy. (4.69)
Equation (4.70) becomes upon substitution of a phase shift k = kx0 x+ ky0 y
∑p,q
f(p, q) =1
2jac
∑p,q
e−jkzpq |z|√k2 −
(2πpa
)2+(
2πqc
)2e−jkxpxe−jkyqy (4.70)
Shanks’ Transform [90, 99, 101, 43]
For alternating series, the partial sums of a sequence can be treated as a mathematical
transient of the sum S of the series. Assume the nth partial sum is
Sn = S +n∑
k=0
αkqnk
for |qk| < 1. Hence, Sn → S as n→ ∞. The sum S can be found by expressing Sn in terms
of S, α, and q for indices (n− 1), n, and (n + 1),
Sn = S + αqn
Sn−1 = S + αq(n−1)
Sn+1 = S + αq(n+1)
96
and solving for the sum S. The sum is found to be equal to
S =Sn+1Sn−1 − S2
n
Sn+1 + Sn−1 − 2Sn. (4.71)
Two simple examples will illustrate that the sum S of simple alternating series can be
determined without summing an inordinate number of terms.
Example 1. Geometric series The geometric series
∞∑k=0
qk = 1 + q + q2 + · · · (4.72)
converges to sum 1/(q − 1) if |q| < 1 and αk = 1k. Setting S0 = 1, S1 = 1 + q, and
S2 = 1 + q + q2, and solving for S using (4.71) yields
=(1 + q + q2)− (1 + q)2
1 + q + q2 + 1− 2(1 + q)=
−q−q + q2
=1
q − 1.
Example 2. Alternating geometric series
∞∑k=0
(−1)kqk = 1− q + q2 − · · · (4.73)
converges to sum 1/(q + 1) if |q| < 1 and αk = (−1)k. Setting S0 = 1, S1 = 1 − q, andS2 = 1− q + q2, and solving for S using (4.71) yields
S =(1− q + q2)− (1− q)2
1− q + q2 + 1− 2(1− q) =q
q2 + q
=1
q + 1.
Another effective transform for alternating series is the Shanks’ transform. The algo-
rithm for the Shanks’ transform of a sequence of partial sums is
S = e1(Sn+1) +1
e1(Sn+1)− e1(Sn) , (4.74)
where
e0(Sn) = Sn, e1(Sn) =1
e0(Sn+1)− e0(Sn) .
97
Higher order Shanks transforms can be carried out using Wynn’s ε-algorithm [124, 15, 101]:
es+1(Sn) = es−1(Sn+1) +1
es(Sn+1)− es(Sn) , s = 1, 2, · · · (4.75)
where
e0(Sn) = Sn, e1(Sn) =1
e0(Sn+1)− e0(Sn)Only the even order terms e2r(Sn) are Shanks’ transforms of order r approximating S. The
process is applied continually until a desired criterion is reached. For the series used in this
work, the following convergence criterion was used. The convergence factor εc is defined by∣∣∣∣ek(Sn−k)− ek−2(Sn−k+2)ek(Sn−k)
∣∣∣∣ ≤ εc. (4.76)
In order to assure that the summation has adequately converged, the algorithm is contin-
ued until three successive values of ek+2(Sn−k−2), ek+4(Sn−k−4), and ek+6(Sn−k−6) satisfy
(4.76).
Table 4.3: Convergence of higher order Shanks’ transforms of Leibnitz series as a functionof order and number, π=3.14159265358979 . . .
n e0 = Sn e2 e4 e6 e8 e10
0 4.0000000 3.1666667 3.1423423 3.1416149 3.1415933 3.1415926(7)
1 2.6666667 3.1333333 3.1413919 3.1415873 3.1415925
2 3.4666667 3.1452381 3.1416667 3.1415943 3.1415927
3 2.8952381 3.1396825 3.1415634 3.1415921
4 3.3396825 3.1427128 3.1416065 3.1415967
5 2.9760462 3.1408813 3.1415854
6 3.2837385 3.1420718 3.1415929
7 3.0170718 3.1412548
8 3.2523659 3.1415927
9 3.0418396
10 3.2323158
Example. Leibnitz series
π = 4∞∑n=0
(−1)n
2n+ 1= 3.14159265358979 . . . (4.77)
98
Table 4.3 shows the results of applying various order Shanks’ transforms to Leibnitz series.
One notes that using a fourth-order Shanks’ transform is accurate to six significant digits
but only requires the computation of 10 terms in the series and 40 floating point operations
(FLOPs). A direct summation of the series would require calculating and summing 25×106
terms to achieve a similar accuracy of five significant digits.
The Shanks’ transform is simple and efficient – the first N terms of the series are the
only terms computed. The higher order transforms are applied to the N terms using the
simple algorithm listed above. The total number of floating point operations required to
compute a (N, k) order Shanks’ transform can be calculated to be
FLOPs =k∑
n=0
(N − n) = N(k + 1)−k∑
n=1
n
=(N − k
2
)(k + 1). (4.78)
In Table 4.4, the error and relative computation time for various combination order Shanks’
transformations on (4.77) are shown. The baseline computation time is determined by
Table 4.4: Error for various (N, k) Shanks’ transforms for Leibnitz series
N k ek(SN ) Error FLOPs
10 0 3.23231580 9.07 × 10−2
50 0 3.16119861 1.96 × 10−2
100 0 3.15149340 9.90 × 10−3
10 2 3.14173610 1.43 × 10−4 27
10 4 3.14159331 6.58 × 10−7 40
10 6 3.14159266 5.11 × 10−9 49
10 8 3.14159265 5.43× 10−11 54
50 2 3.14159443 1.78 × 10−6 147
50 4 3.14159265 5.97× 10−10 240
100 2 3.14159443 2.36 × 10−7 297
assuming that 100 terms of the sum are computed. The relative computational time for
the various order transforms are computed by applying the transforms to different num-
bers of terms in the series. Because the determination of the higher order transforms are
computationally inexpensive, it is efficient to set k=N -2. When k=N -2, the error can be
minimized without increasing the computational cost significantly. The total number of
99
FLOPs required for a (N ,N -2) order Shanks’ transform is
FLOPs =N−2∑n=0
(N − n) =(N
2+ 1)(N − 1) N2
2(4.79)
For a double-sided series, the number of FLOPs required is simply twice the number for a
single-sided series and can be approximated byN2. For example, a (10,4) Shanks’ transform
of (4.77) results in an error of 6.58×10−7 and requires 80 FLOPs. However, simply carrying
out two more transforms (requiring 28 more FLOPs) yields an error of only 5.43 × 10−11.
Also, note that a (10,8) transform yields a significantly smaller error than a (100,2) trans-
form and does so at a significantly reduced computational cost. For more realistic series
where calculating the individual terms is expensive, the determination of the minimum num-
ber of terms N and the subsequent order k=N -2 of the transform to accurately compute
the sum is imperative.
(30,28) Shanks’ transform of inner sum of Zyymm impedance matrix element
An example of the significant increase in accuracy and decrease in computational cost that
implementing a (N ,N -2) Shanks’ transform produces is easily seen by computing the inner
sum for the (m,n) element of the Zyy submatrix in (4.41) and implementing a (30,28)
Shanks’ transform. In a double summation, such as those found in this work, the Shanks’
transform is applied to the sequence of partial sums over the inner indices for a specific
outer index. The transform is then applied to the outer sequence. For convenience, the Zyy
submatrix is repeated below
Zyymn = −k0Z0∆x∆y
2ac
∑p
∑q
(1− k2
yq
k20
)
×[2j − 2je−jkzpq ∆z/2
k2zpq
+Ry(γpq)∆z sinc(kzpq∆z
2
)e−j2kzpqzm
kzpq
]
× sinc(kxp∆x
2
)sinc
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn) (4.80)
The slowly converging part of the sum in (4.80) is the first expression. Simplifying (4.80)
by assuming that Ry = 0, xm = xn, ym = yn, and zm = zn yields an equation whose
convergence rate is relatively unchanged by the simplification. The resulting equation is
Zyy = −2k0Z0
ac
∑p
sin(kxp∆x/2
)kxp
∑q
(1− k2
yq
k20
)sin(kyq∆y/2
)kyq
[2j − 2je−jkzpq∆z/2
k2zpq
].
(4.81)
100
FLOPs
102 103 104 105 106 107
Rel
ativ
e er
ror
10-8
10-7
10-6
10-5
10-4
10-3
10-2Direct sumShanks (12,10)Shanks (16,14)Shanks (20,18)Shanks (24,22)Shanks (28,26)
Figure 4.7: Relative error as a function of floating point operations for computing variousorder Shanks’ transforms and the direct sum of the inner sum of (4.80)
The inner sum is evaluated assuming the index of the outer sum p = 0. It is clear that
implementing a (N ,N -2) Shanks’ transform to sum the series minimizes the relative error for
a given computational cost. The relative error as a function of computational cost (FLOPs)
of implementing various order Shanks’ transforms is shown in Figure 4.7. The number of
floating point operations for the direct sum of the series is also included for comparison.
Notice the dramatic decrease in the relative error of the sum for a given number of FLOPS
required to sum the inner sum of (4.80) to a given accuracy by implementing different
(N ,N -2) Shanks’ transforms. To achieve a relative error of 10−4 for the inner sum, only
103 FLOPS are required by the Shanks’ transform, whereas 104 FLOPS are needed by the
direct sum. The disparity in computational cost between computing the direct sum and
implementing a Shanks’ transform is even more significant for a relative error of less than
10−6.
If (4.80) is used without simplification, the series that must be summed is
Zyy = −2k0Z0
ac
∑p
sin(kxp∆x/2
)kxp
e−jkxp(xm−xn)
×∑q
(1− k2
yq
k20
)sin(kyq∆y/2
)kyq
e−jkyq (ym−yn)
[2j − 2je−jkzpq ∆z/2
k2zpq
](4.82)
101
However, the same Shanks’ transform is used without modification to sum this element.
This is significant in light of the fact that all of the sums can be computed using the same
general subroutine. This is not true for other summation acceleration techniques.
Kummer’s Method [51, 99]
The statement that the rate of convergence of a series is determined by the asymptotic
form of the series is the basis for Kummer’s method. Suppose a series f(n) has an asymptotic
form f∞(n). An equivalent expression to f(n) can be found if the asymptotic series is
subtracted from and added back to the original series as seen below∞∑
n=−∞f(n) =
∞∑n=−∞
[f(n)− f∞(n)] +∞∑
n=−∞f∞(n). (4.83)
Usually, f∞(n) is chosen in such a way that the last series has a known closed-form expres-
sion. For complicated series, however, obtaining a closed-form expression for the asymptotic
series can be a tedious task.
The double summation that is found in this work is written symbolically as
S =∞∑m=0
∞∑n=0
Smn.
The asymptotic series Amn is subtracted from and added to the original series Smn
S =∞∑m=0
∞∑n=0
(Smn −Amn) +∞∑n=0
Amn
.
If the asymptotic series Amn has a closed-form solution in terms of the inner index n equal
to Am, then the above equation can be written as
S =∞∑m=0
∞∑n=0
(Smn −Amn) +Am.
The sum Smn −Amn converges much more rapidly than the direct sum of Smn.
Additional series acceleration can be achieved for some series by incorporating a similar
technique on the tail contribution of the series. Substituting Zmn = Smn−Amn and writing
the series Zmn as the sum of (N + 1) terms and the remaining tail contribution yields
S =∞∑m=0
N−1∑n=0
Zmn +∞∑
n=N
Zmn +Am
.
The asymptotic tail series Tmn of the tail contribution is subtracted from and added to the
original series Zmn much like before
S =∞∑m=0
N−1∑n=0
Zmn +∞∑
n=N
(Zmn − Tmn) +∞∑
n=N
Tmn +Am
.
102
The tail contribution is now written as∑∞
n=N Tmn =∑∞
n=0 Tmn−∑N−1
n=0 Tmn and the above
equation can be written as
S =∞∑m=0
N−1∑n=0
Zmn +∞∑
n=N
(Zmn − Tmn) +∞∑n=0
Tmn −N−1∑n=0
Tmn +Am
.
If Tm is the closed-form solution for the inner sum of Tmn, then
S =∞∑m=0
N−1∑n=0
Zmn +∞∑
n=N
(Zmn − Tmn) + Tm −N−1∑n=0
Tmn +Am
.
The sum Zmn − Tmn also converges much more rapidly than the direct sum of Zmn.
Kummer’s method applied to inner sum of Zyymm
An example of the usefulness of Kummer’s method is seen in computing the same inner sum
for the (m,n) element of the Zyy submatrix of (4.41) as was determined using a Shanks’
transform. Beginning with (4.80), the inner sum is computed using Kummer’s method.
Only one expression of the inner series is not exponentially convergent:
S′p = 2j
∑q
(1− k2
yq
k20
)sin(kyq∆y/2
)kyqk
2zpq
(4.84)
where S′p is the sum over all q excluding the following q = 0 term
Sp0 =
(1− k2
y0
k20
)sin (ky0∆y/2)
ky0
[2j − 2je−jkzp0∆z/2
k2zp0
].
For large q, the sum in (4.84) can be approximated by
S′p = −2j
k20
′∑q
2πqc
1
k20 − k2
xp−(
2πqc
)2 sin(kyq∆y/2
)
=2jk20
′∑q
2πqc
( c2π
)2 1
q2 −[(
ck02π
)2 −(ckxp
2π
)2] sin (kyq∆y/2
)(4.85)
with
kyq 2πqc, k2
zpq k2
0 − k2xp
−(2πqc
)2
.
If we represent the denominator as q2 − α2p, where α2
p = (ck0/2π)2 − (ckxp/2π
)2, thenS′p =
cj
πk20
′∑q
q
q2 − α2p
sin(kyq∆y/2
). (4.86)
103
Noting that kyq = ky0 +2πqc and using a simple trigonometric identity, (4.86) can be written
as
S′p =
cj
πk20
′∑q
q
q2 − α2p
[sin (ky0∆y/2) cos (qπ∆y/c) + cos (ky0∆y/2) sin (qπ∆y/c)
]. (4.87)
The sums in (4.87) have asymptotic forms that can be evaluated analytically. If we rewrite∑′q as
∑−1q=−∞+
∑∞q=1 and observe that
q
q2 − α2p
cos qζ = − q
q2 − α2p
cos(−qζ)q
q2 − α2p
sin qζ =q
q2 − α2p
sin(−qζ)
then (4.87) can be rewritten as
S′p =
2cjπk2
0
cos (ky0∆y/2)∞∑q=1
q
q2 − α2p
sin qζ (4.89)
where ζ = π∆y/c. The sum in (4.89) has an exact solution given by
∞∑q=1
q sin qζq2 + α2
p
=π
2sinhαp(π − ζ)
sinhαpπ, α2
p > 0, 0 < ζ < 2π (4.90a)
∞∑q=1
q sin qζq2 − α2
p
=π
2sinαp(π − ζ)
sinαpπ, α2
p > 0, 0 < ζ < 2π. (4.90b)
The asymptotic value S∞p of (4.86) is
S∞p =
cj
k20
cos (ky0∆y/2)sinαp(π − ζ)
sinαpπ.
Since Sp = Sp0 + S′p + S∞
p ,
Sp =∑q
(1− k2
yq
k20
)sin(kyq∆y/2
)kyq
[2j − 2je−jkzpq ∆z/2
k2zpq
]+cj
k20
cos (ky0∆y/2)sinαp(π − ζ)
sinαpπ
− 2cjπk2
0
cos (ky0∆y/2)∞∑q=1
q
q2 − α2p
sin qζ (4.91)
As can be seen in Figure 4.8, this method also significantly reduces the relative error for a
given computational cost. Included for reference is the Shanks’ transform and direct sum
data of Figure 4.7. A similar convergence rate as the Shanks’ transform is found using
Kummer’s method. However, even this simple series requires concentrated analytical effort.
104
FLOPs
102 103 104 105 106 107
Rel
ativ
e er
ror
10-8
10-7
10-6
10-5
10-4
10-3
10-2Direct sumShanks (12,10)Shanks (16,14)Shanks (20,18)Shanks (24,22)Shanks (28,26)Kummer
Figure 4.8: Relative error as a function of floating point operations for computing variousorder Shanks’ transforms, Kummer’s transforms, and the direct sum of the innersum of (4.80)
If (4.41) is used without simplification, i.e. Ry = 0, xm = xn, ym = yn, but zm = zn,
the series that must be summed is
Zyy = −2k0Z0
ac
∑p
sin(kxp∆x/2
)kxp
e−jkxp(xm−xn)
×∑q
(1− k2
yq
k20
)sin(kyq∆y/2
)kyq
e−jkyq (ym−yn)
[2j − 2je−jkzpq ∆z/2
k2zpq
]. (4.92)
Implementing the above sum using a Shanks’ transform is easily done by applying the
general subroutine to the first (N +1) terms of the series. However, a new analytic solution
must be determined when implementing Kummer’s method. The term of the inner series
that is not exponentially convergent is
S′p =
2jk20
∑q
(k20 − k2
yq
)kyqk
2zpq
sin(kyq∆y/2
)e−jkyq (ym−yn) (4.93)
where S′p is the sum over all q excluding the following q = 0 term
Sp0 =
(1− k2
y0
k20
)sin (ky0∆y/2)
ky0e−jky0(ym−yn)
[2j − 2je−jkzp0∆z/2
k2zp0
].
105
For large q, the sum in (4.93) can be approximated by
S′p = −2j
k20
′∑q
2πqc
1
k20 − k2
xp−(
2πqc
)2 sin(kyq∆y/2
)e−jkyq (ym−yn)
=2jk20
′∑q
2πqc
( c2π
)2 1
q2 −[(
ckxp
2π
)2 −(ck02π
)2] sin (kyq∆y/2
)e−jkyq (ym−yn) (4.94)
with
kyq 2πqc, k2
zpq k2
0 − k2xp
−(2πqc
)2
.
If we represent the denominator as q2 − α2p, where α2
p = (ck0/2π)2 − (ckxp/2π
)2, thenS′p =
cj
πk20
′∑q
q
q2 − α2p
sin(kyq∆y/2
)e−jkyq (ym−yn) (4.95)
Similar to before, (4.95) can be written as
S′p =
cj
πk20
e−jky0(ym−yn) sin (ky0∆y/2)′∑q
q
q2 − α2p
cos (qπ∆y/c) e−j2πq
c(ym−yn)
+cj
πk20
e−jky0(ym−yn) cos (ky0∆y/2)′∑q
q
q2 − α2p
sin (qπ∆y/c) e−j2πq
c(ym−yn). (4.96)
The sums in (4.96) have asymptotic forms that can be evaluated analytically with a little
analytical effort. Equation (4.96) can be rewritten as
S′p =
cj
πk20
e−jky0(ym−yn) sin (ky0∆y/2)∞∑q=1
q
q2 − α2p
[−j sin q(ζ+) + j sin q(ζ−)
]
+cj
πk20
e−jky0(ym−yn) cos (ky0∆y/2)∞∑q=1
q
q2 − α2p
[sin q(ζ+) + sin q(ζ−)
](4.97)
where ζ± = π∆y [1± 2(m− n)] /2. Using (4.90), the asymptotic value of (4.97) can be
written as
S∞p =
c
2k20
e−jky0(ym−yn) sin (ky0∆y/2)[sinαp(π − ζ+)
sinαpπ− sinαp(π − ζ−)
sinαpπ
]
+cj
2k20
e−jky0(ym−yn) cos (ky0∆y/2)[sinαp(π − ζ+)
sinαpπ+
sinαp(π − ζ−)sinαpπ
]. (4.98)
106
Again, Sp = Sp0 + S∞p + S′
p,
Sp =∑q
(1− k2
yq
k20
)sin(kyq∆y/2
)kyq
e−jkyq (ym−yn)
[2j − 2je−jkzpq ∆z/2
k2zpq
]
− cj
πk20
e−jky0(ym−yn) sin (ky0∆y/2)∞∑q=1
q
q2 − α2p
[−j sin q(ζ+) + j sin q(ζ−)
]
− cj
πk20
e−jky0(ym−yn) cos (ky0∆y/2)∞∑q=1
q
q2 − α2p
[sin q(ζ+) + sin q(ζ−)
]
+c
2k20
e−jky0(ym−yn) sin (ky0∆y/2)[sinαp(π − ζ+)
sinαpπ− sinαp(π − ζ−)
sinαpπ
]
+cj
2k20
e−jky0(ym−yn) cos (ky0∆y/2)[sinαp(π − ζ+)
sinαpπ+
sinαp(π − ζ−)sinαpπ
](4.99)
As can be clearly seen, the analytical effort to determine the asymptotic form of the elements
in the impedance matrix is not trivial; the requisite computer coding to implement this
solution lends itself to difficult debugging. Additionally, applying Kummer’s method to the
inner sum does not produce an analytical solution for the outer sum. Thus, one must either
compute an addition analytical solution for the outer sum or find another mechanism to
compute the sum. In light of the difficulties found in computing realizable double sums,
implementing a general Shanks’ transform may be more profitable.
Another application of Kummer’s method in the acceleration of the PFSGF series is
to simply sum the series incorporating a “smoothing” or “acceleration” parameter in the
propagation constant. The asymptotic form of the new “modified” Green’s function is then
subtracted from the original series and the transformed series is added back [40].
4.5.2 Acceleration Techniques Used Elsewhere
Ewald Transformation [30, 39, 65, 27]
The Ewald transformation expresses the periodic free-space Green’s function as the sum
of two doubly infinite series, namely one series summed in the spectral domain and one series
summed in the spatial domain. Similarly to Kummer’s method, the Ewald transformation
casts the PFSGF into a hybrid form more amenable to acceleration. The Ewald transfor-
mation results in series that utilizes the complementary error function yielding two more
rapidly converging series.
107
ρ-algorithm [124, 102]
For monotonic series, the ρ-algorithm is a very rapidly converging series acceleration
technique. However, it does not fare as well with alternating series. The simple ρ-algorithm
can be computed as follows
ρ(n)k = ρ(n+1)
k−2 +k
ρ(n+1)k−1 − ρ(n)
k−1
, k = 1, 2, · · · (4.100)
where
ρ(n)−1 = 0, ρ
(n)0 = Sn (4.101)
and where k is the order of the algorithm. The even order terms, ρ(n)2k , give the estimate of
the sum.
θ-algorithm [15, 104]
Another rapidly accelerating series technique for alternating series is the θ-algorithm.
The θ-algorithm can be computed as follows
θ(n)2k+2 = θ(n+1)
2k +
[θ(n+2)2k − θ(n+1)
2k
] [θ(n+2)2k+1 − θ(n+1)
2k+1
][θ(n+2)2k+1 − 2θ(n+1)
2k+1 + θ(n)2k+1
] , k = 0, 1, 2, · · · (4.102)
and the odd order terms by
θ(n)2k+1 = θ(n+1)
2k−1 +1[
θ(n+1)2k − θ(n)
2k
] , k = 1, 2, · · · (4.103)
where
θ(n)−1 = 0, θ
(n)0 = Sn (4.104)
The even order terms, θ(n)2k+2, give the estimate of S. Like the ρ-algorithm, the θ-algorithm
is easy to implement.
108
Chebyshev-Toeplitz algorithm [121, 105]
The Chebyshev-Toeplitz (CT) algorithm can be computed as follows
t(n)−1 = 0, t
(n)0 = Sn, σ0 = 1, (4.105)
t(n)1 = t(n)
0 + 2t(n+1)0 , σ1 = 3, (4.106)
t(n)k+1 = 2t(n)
k + 4t(n+1)k − t(n)
k−1, k = 1, 2, · · · (4.107)
σk+1 = 6σk − σk−1, k = 1, 2, · · · (4.108)
T(n)k =
t(n)k
σk, k = 0, 1, 2, · · · (4.109)
The nth iterate of the CT algorithm is given by T (n)k which gives as estimate of the sum of
the series.
Levin t-transform [54, 106]
The t-transform algorithm can be computed as follows
t(n)k =
∑ki=0(−1)i
(ki
)(n+in+k
)(k−1)(
Sn+i
Sn+i−1−Sn+i
)∑k
i=0(−1)i(ki
)(n+in+k
)(k−1)(
1Sn+i−1−Sn+i
) , k = 1, 2, · · · (4.110)
The nth iterate of the Levin t-transform is given by T (n)k which gives an estimate of sum of
the series.
4.6 Conclusions
A full-wave IE/MoM solution that determines the general three-dimensional scattering
from an inhomogeneous doubly periodic layer above a half-space medium has been for-
mulated, implemented, and validated. Equivalent volume polarization currents are used
to replace the dielectric material and the combination of a Poisson transformation and a
Shanks’ transformation is used to improve the speed and accuracy of the impedance matrix
element computations. The accuracy of the solution is primarily dependent on three fac-
tors: (i.) representing the dielectric material with an appropriate number of subsectional
unknowns, (ii.) including enough terms in the resulting Floquet summations, and (iii.)
the choice of an appropriate series acceleration convergence technique. A general rule-of-
thumb in MoM formulations is to use 5–10 subsectional unknowns per half wavelength in
the medium. This is the minimum number that should be used when modeling the z varia-
tion of the material. Fewer unknowns per wavelength are required to model the transverse
109
variation to obtain a given accuracy because the fields are more smoothly varying in the
transverse direction.
The solution procedure is valid for reasonable combinations of unit cell spacing, material
shape, dielectric constant, and number of layers. This solution procedure is subsequently
used in Chapter 5 to validate the modeling of an equivalent uniaxial layer that replaces the
doubly periodic layer for use in rectangular patch antenna applications.
110
CHAPTER 5
Radiation Properties of a Rectangular Microstrip Patch on a
Uniaxial Substrate
5.1 Introduction
Of interest to the applied microwave community is the implementation of a microstrip
patch antenna on a doubly periodic dielectric layered medium. Preliminary investigations
of simple Hertzian dipoles radiating over this structure have been addressed by Yang [133]
using an analytic array scanning technique [67]. However, the solution in [133] is compu-
tationally intensive and has not been extended for arbitrary polarization. A simpler (and
perhaps more insightful) solution where the doubly periodic dielectric layer is emulated by
an equivalent anisotropic (uniaxial) material is developed in the following sections. The
uniaxial model for the doubly periodic dielectric layer is validated using plane wave reflec-
tion coefficients for a variety of filling fractions, angles, and permittivities. Although the
solution of a similar approximation has been developed for two-dimensional scattering from
periodicities in one-dimension [85, 87], this is the first known treatment of such a solution
for full three-dimensional scattering from dielectric periodicities in two directions.
Traditionally, microstrip patch antennas have been integrated on relatively low permit-
tivity substrates in order to improve antenna performance. Since microstrip patches are
effectively a radiating resonator, the large Q produced by confining the stored energy in a
thin region under the patch necessarily narrows the bandwidth significantly. Typical band-
widths for traditional patch designs are on the order of 2–4%. New designs for microstrip
patch feeds have increased bandwidths but at the cost of increased complexity. Integrating
the antenna on higher permittivity substrates is preferred to minimize circuit size and spu-
rious radiation [75, 89] but at the cost of confining the potential radiating energy even more
111
tightly. This trade-off between good antenna performance and good circuit performance is
a key design feature found in many microstrip antenna designs. The compromise can be
achieved primarily through the proper design of any or all of the three main components in a
microstrip structure: radiating element, feed mechanism, and substrate choice. (Extensive
bibliographies can be found in [75, 89].) The focus of this chapter is the choice and design of
an effective anisotropic substrate for use in microstrip antenna applications. Additionally,
the solution for a rectangular patch element radiating over the equivalent uniaxial substrate
is carried out accurately and efficiently using Ansoft’s High Frequency Structure Simulator
that incorporates anisotropic substrates.
In order to reduce the often unwanted formation of surface waves which can lead to pat-
tern degradation and low efficiencies, relatively thin homogeneous substrates (thicknesses
less than 0.05λ0) are used. However, as noted before, this limitation has severe conse-
quences. Although a good deal of attention has focused on integrating microstrip patches
on homogeneous substrates, many of the practical substrates in use today such as sapphire,
Epsilam-10, and boron nitride, have a significant amount of (uniaxial) anisotropy. The
uniaxial materials have two effective homogeneous permittivities: one permittivity aligned
parallel to the optical axis of the material and one permittivity aligned perpendicular to the
optical axis [47, 85, 87]. Rigorous full-wave solutions, albeit computationally intensive ones,
have been developed to characterize the effect of anisotropy on various patch antenna param-
eters such as resonant length and efficiency [74, 68, 131]. Other investigations [19, 14, 123]
have focused on supplementing the conclusions drawn in earlier investigations, namely, that
the primary effect of anisotropy on rectangular patch antennas is the change in its resonant
length. This is significant because of the narrow bandwidth of the patch itself. The rel-
atively large shift in resonant frequency produced in many of the modern substrates may
actually force a rectangular patch designed to operate at a specific frequency to radiate out-
side of the antenna bandwidth [74]. Additionally, anisotropic effects are found that shape
the radiation pattern of the patch and thus in an array configuration, the coupling to other
elements.
Because uniaxial substrates are often expensive to manufacture and have limited flexibil-
ity for design, uniaxial substrates can be emulated (and easily fabricated) by incorporating
periodic inclusions in an otherwise homogeneous substrate. In particular, if the wavelength
is sufficiently large compared with the period of the array, the doubly periodic dielectric layer
(in the transverse direction) can emulate a uniaxial substrate with an optical axis aligned
in a direction normal to the layer [85, 87]. The periodic structure shown in Figure 5.1(a)
112
can be easily constructed using simple milling or etching techniques from simple inexpen-
ε
ε
ε0
2
1
(a)
ε
ε
ε0
2
t , εz
(b)
Figure 5.1: (a) Periodic dielectric layer over a layered medium and (b) equivalent uniaxialmedium with transverse permittivity εt and axial permittivity εz over a layeredmedium
sive, homogeneous substrates. This is significant because some common uniaxial materials
such as sapphire that are expensive to grow [6] can be “artificially” replicated easily and
inexpensively. Additionally, the artificial nature of the periodic uniaxial substrate per-
mits the creative design of new substrates with the expanded freedom of anisotropic ratio,
background permittivity, and/or fabrication technique.
5.2 Equivalent Uniaxial Modeling
The equivalent uniaxial material of Figure 5.1(b) with the optical axis aligned with the
z axis has an assumed permittivity tensor of
¯ε = ε0
εx 0 0
0 εy 0
0 0 εz
(5.1)
where εx=εy=εt is the transverse component and εz is the component along the direction
of the optical axis.
Two of the well-known dielectric mixing models1 for two-phase media (i.e., ε is piecewise
constant and takes on only two distinct values each) are the Hashin-Shtrikman bounds [35,
47] and the Licktenecker bounds [25]. As with many of the early approximations, these
solutions are determined using variational methods, and thus, the solutions take on the
form of bounds for the equivalent two-phase materials (permittivities, conductivities, or1These models belong to a more general class of dielectric mixing formulas that include the Maxwell-
Garnet mixing formula (accurate for relatively low volume fractions) and the Polder van Santen mixingformula (accurate for relatively low permittivity changes).
113
permeabilities). For the Hashin-Shtrikman and the Licktenecker models outlined below,
both the unit cell and the material inclusions are assumed to have square2 cross-sections.
5.2.1 Hashin-Shtrikman Model
A good approximation (assuming the inclusions are small with respect to wavelength) for
the longitudinal component of the permittivity tensor, εz, is simply to take the volumetric
averages of the constituent materials and is given by
εz = (1− f2)εa + f2εb (5.2)
where f is the filling fraction, εa is the background permittivity, and εb is the permittivity of
the material blocks or inclusions. The generalized Hashin-Shtrikman formulas provide upper
and lower bounds for the transverse component of the permittivity tensor, εt [48, 47]. While
no known closed form expression exists for the transverse permittivity, evidence suggests
that the lower and upper Hashin-Shtrikman bounds provide good accuracy to approximate
the value over a wide range of parameters for many practical substrates. The lower and
upper bounds, denoted εLHS and εUHS , respectively, are defined as
εLHS = εa2εb + f2 (εa − εb)2εa + f2 (εb − εa) (5.3a)
εUHS = εb(εa + εb) + f2 (εa − εb)(εa + εb) + f2 (εb − εa) . (5.3b)
5.2.2 Licktenecker Model
Another set of bounds that serves to approximate the effective transverse permittivity
can be derived using the Licktenecker bounds [25]. The Licktenecker bounds can be written
as ∫ b
0
dy∫ a0
dxε(x,y)
≤ εx ≤∫ a
0
dx∫ b0
dyε(x,y)
(5.4a)
∫ a
0
dx∫ b0
dyε(x,y)
≤ εy ≤∫ b
0
dy∫ a0
dxε(x,y)
. (5.4b)
However, if the inclusion is symmetric about the line x=y, the bounds take on simpler
forms since εx=εy=εt. Carrying out the integrals in (5.4), the lower and upper Licktenecker2Similar results can be derived for circular inclusions in a square unit cell but are not presented. For
electrically small cells, the particular geometric shape is insignificant compared to the volumetric averagesof the constituent phases.
114
bounds, denoted εLL and εUL , respectively, are determined to be
εLL =fεb + (1− f)εa
fεa + f(1− f)εb + (1− f)2εa (5.5a)
εUL =fεb + f(1− f)εa + (1− f)2εb
fεa + (1− f)εb (5.5b)
where f is the filling fraction b/a, εa is the background permittivity, and εb is the permittiv-
ity of the material blocks or inclusions. Note that if the phasing of the material is reversed,
1/εLL = εUL .
The effective transverse permittivity of the two-dimensional periodic media determined
by the Hashin-Shtrikman and Licktenecker bounds is graphed as a function of filling fraction
b/a for εb = 10εa in Figure 5.2. Also included in the figure is the solution obtained from
f
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
ε t
1
2
3
4
5
6
7
8
9
10εz
εHSL
εHSU
εLL
εLU
Figure 5.2: Effective transverse permittivity as a function of filling fraction for εb=10εa
simply weighting the constitutive phases by their respective volume fractions. Note that
this is nothing more than the effective permittivity in the axial direction determined from
(5.2). Consequently, the assumption that the permittivity of the two dimensional periodic
substrate (even for electrically small cells) has an effective homogeneous permittivity leads
to erroneous results. Additionally, one observes that the upper Hashin-Shtrikman bound
is grossly in error. However, this error has been reported previously in the literature [47].
The other three bounds are reasonably precise and have been reported to have small errors
of less than 4%. The solutions obtained for the Licktenecker bounds are precise, and since
they are derived using specific information about the geometry of the inclusions, one can
115
conclude these bounds are more accurate. Overall, the precision of the three remaining
bounds is found to increase with increasing filling fraction.
f
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
ε t
1
2
3
4
5
6
7
8
9
10εz
εHSL
εHSU
εLL
εLU
Figure 5.3: Effective transverse permittivity as a function of filling fraction for εa=10εb
If the phases of the two materials are reversed as shown in Figure 5.3, a noticeable
change in precision can be seen. The upper Hashin-Shtrikman bound is again seen to have
significant error. The remaining three bounds are reasonably close for large filling fractions,
but have some noticeable differences at lower filling fractions. Again, the incorporation
of geometrical features necessarily makes the Licktenecker bounds a more accurate solu-
tion. Consequently, a good approximation for the effective transverse permittivity can be
determined by taking the geometric mean of εLL and εUL
εt =√εLLε
UL . (5.6)
Not only are the values obtained using the geometric mean of εLL and εUL remarkably close
to the values yielded by the lower Hashin-Shtrikman bound for the phase set in Figure 5.2,
but also for the second case shown in Figure 5.3 when the phases are reversed.
An interesting feature of the emulated anisotropic material is the ability to achieve a
desired anisotropic ratio (AR) by incorporating either a periodic dielectric region in an air
background or by including periodic air inclusions in a dielectric background depending on
the application and ease of fabrication. The AR is a function of the dielectric contrast
between the material blocks or inclusions and the background medium. If the effective
116
transverse permittivity εt is greater than the effective axial permittivity εz (εt ≥ εz), the
material is said to have negative uniaxial anisotropy (AR > 1). Conversely, if the effec-
tive axial permittivity εz is greater than the effective transverse permittivity εt (εt ≤ εz),
the material is said to have positive uniaxial anisotropy (AR < 1). For example, in the
microwave region, sapphire (a common substrate material used in microwave integrated
circuits) has an AR of 0.81 (εt=9.4, εz=11.6) [6]. Thus, sapphire has positive uniaxial
anisotropy. Boron nitride has an AR of 1.5 (εt=5.12, εz=3.4) and Epsilam-10 has an AR
of 1.26 (εt=13.0, εz=10.3) [6]. Hence, boron nitride and Epsilam-10 have negative uniaxial
anisotropy. Most of the soft composite ceramic-impregnated teflon-type substrates, such
as Rogers’ RT/Duroid 6000 series, have anisotropic ratios greater than one. The physical
significance of the AR will be addressed in the next section.
For the substrates with periodicities in the plane of the substrate, the emulated material
is necessarily a positive uniaxial material. Since εt and εz are known quantities, specifying
one of the three unknowns, namely εa, εb or f , one can solve for the remaining two unknowns
of interest from (5.2) and (5.5). For example, sapphire can be emulated by a medium
consisting of periodic dielectric rods with a dielectric constant of 12.1 and filling fraction
equal to 0.95, or by a medium consisting of periodic air inclusions in a dielectric background
with dielectric constant of 16.0 and filling fraction equal to 0.3 as seen in Figure 5.4. The
second medium can be viewed more intuitively as doubly periodic set of dielectric veins
a
b
ε2ε1
b = 0.95 aε = 12.1 ε2 1
(a)
a
b
ε2
ε1
b = 0.30 aε = 16.0 ε2 1
(b)
Figure 5.4: Uniaxial nature of sapphire modeled by (a) periodic dielectric rods withε2=12.1ε1 and b/a=0.95 and (b) periodic dielectric veins with ε2=16.0ε1 andb/a=0.30
117
in an air background. This observation is consistent with the two-dimensional periodic
structures developed in Chapter 3 and detailed in Section 3.2.2.
As was introduced earlier, the most significant application for the emulated uniaxial
materials is the variety of artificial substrate designs available to the RF designer. Naturally
occurring uniaxial materials have fixed anisotropic ratios and permittivity tensors and limit
important parameter choices. However, depending on the application and the materials
available for use, a particular AR can be be emulated. This flexibility allows the RF designer
not only the choose of a particular AR, but also the choice of the dielectric constant of the
background medium and/or inclusions.
Using the Hashin-Shtrikman or Licktenecker bounds, one can compute the minimum
achievable AR of the equivalent positive uniaxial medium by differentiating the ratio εt/εz
with respect to the filling fraction f where εt is found using (5.3b) or (5.5) and εz is found
from (5.2). For electrically small cells, the minimum AR that can be achieved for a given
dielectric constant ratio, either for εb/εa > 1 or εb/εa < 1, is shown in Table 5.1. Note
Table 5.1: Minimum achievable anisotropic ratio (AR)
(εb/εa) Min AR εt εz b/a
1.00 1.00 1.00 1.00 1.00
2.32 0.90 1.45 1.61 0.46
4.00 0.74 1.74 2.35 0.45
6.15 0.59 1.94 3.28 0.44
10.20 0.43 2.13 5.00 0.44
(εa/εb) Min AR εt εz b/a
1.00 1.00 1.00 1.00 1.00
2.32 0.93 1.48 1.58 0.56
4.00 0.85 1.87 2.18 0.61
6.15 0.79 2.23 2.81 0.65
10.20 0.73 2.76 3.78 0.70
that the minimum achievable AR is strongly dependent on the permittivity of the dielectric
contrast between the phases. Thus, in order to emulate a given permittivity set (εt,εz),
the ratio of the relative dielectric constant of the two phases must be large enough to both
produce the minimum achievable AR listed in Table 5.1 and to also achieve the value of
118
the desired axial component of the permittivity tensor. As will be seen in the next section,
the AR is a parameter that has a significant effect on the resonant length of a radiating
structure located on or near such a material.
In order to verify that using the previous models are valid approximations for the uni-
axial layer, the reflection coefficient for the equivalent uniaxial medium is compared to the
IE/MoM numerical solution of the reflection coefficient obtained from the doubly periodic
grounded dielectric layer modeled using polarization currents. Thus, the exact horizontal
and vertical reflection coefficients at the surface of the grounded uniaxial dielectric layer of
thickness t with dielectric permittivity tensor given in (5.1) must be found.
5.3 Plane Wave Reflection Coefficients for a Grounded Uni-
axial Layer
Assuming the equivalent uniaxial layer has the permittivity tensor ¯ε given in (5.1) and
and starting from Maxwell’s equations for a uniaxial medium
∇× E = −jωµH (5.7)
∇× H = jω ¯ε ·E, (5.8)
one can solve for the wave equations that must be satisfied in the medium. Following [74]
the wave equation for the z -component of the electric field can be determined to be
∂2
∂x2Ez +
∂2
∂y2Ez +
εzεt
∂2
∂z2Ez + εzk2
0Ez = 0 (5.9)
where k0=ω√µ0ε0=2π/λ0. The wave equation for the z -component of the magnetic field
can also be determined and is found to be
∂2
∂x2Hz +
∂2
∂y2Hz +
∂2
∂z2Hz + εtk2
0Ez = 0. (5.10)
The dispersion relations for (5.9) and (5.10) can be found by assuming waves of the form
e±jkxxe±jkyye±jkzz. Thus, for TEz (transverse electric to z ) or horizontal polarized waves,
k2z = εtk2
0 − β2 = k2a (5.11)
where β2 = k2x + k2
y and ka is the z -component of the propagation vector. For TMz (trans-
verse magnetic to z ) or vertical polarized waves,
k2z = εtk2
0 − εtεzβ2 = k2
b (5.12)
119
where kb is the z -component of the propagation vector. Note that the uniaxial material
produces two different propagation vectors in the medium for the two polarizations.
We now define the Fourier transform pair
E(x, y, z) =∫∫∞
E(kx, ky, z)ejkzzejkyy dkx dky (5.13a)
E(kx, ky, z) =1
4π2
∫∫∞
E(x, y, z)e−jkzze−jkyy dx dy. (5.13b)
In the transform domain, the transverse fields can be written in terms of Ez and Hz as(εtk
20 +
∂2
∂z2
)Ex = jkx
∂
∂zEz + ωµ0kyHz (5.14a)(
εtk20 +
∂2
∂z2
)Ey = jky
∂
∂zEz − ωµ0kxHz (5.14b)(
εtk20 +
∂2
∂z2
)Hx = jkx
∂
∂zHz − ωε0εtkyEz (5.14c)(
εtk20 +
∂2
∂z2
)Hy = jky
∂
∂zHz + ωε0εtkxEz (5.14d)
where ∂2
∂z2= −k2
z and kz is either ka for TEz or kb for TMz waves. Using the dispersion
relations in (5.11) and (5.12), the above equations can be simplified to
Ex =jkxεzεtβ2
∂
∂zEz +
ωµ0kyβ2
Hz (5.15a)
Ey =jkyεzεtβ2
∂
∂zEz − ωµ0kx
β2Hz (5.15b)
Hx =jkxβ2
∂
∂zHz − ωε0εzky
β2Ez (5.15c)
Hy =jkyβ2
∂
∂zHz +
ωε0εzkxβ2
Ez. (5.15d)
Extending [74], if we assume solutions for Ez and Hz in region 1 have the form
Ez = ejkz0z +Ae−jkz0z (5.16a)
Hz = ejkz0z +Be−jkz0z (5.16b)
and in region 2 have the form
Ez = C cos kbz +D sin kbz (5.17a)
Hz = E sin kaz + F cos kaz, (5.17b)
and apply the appropriate boundary conditions, the solution for the reflection coefficients
for horizontal (A) and vertical (B) polarized electric fields can be determined. The exact
120
horizontal and vertical reflection coefficients at the surface of the grounded uniaxial dielectric
layer of thickness t with dielectric permittivity tensor given in (5.1) are
Rh = A =kz0 sin kat+ jka cos katkz0 sin kat− jka cos kat
ej2kz0t (5.18a)
Rv = B =εtkz0 cos kbt− jkb sin kbtεtkz0 cos kbt+ jkb sin kbt
ej2kz0t (5.18b)
where kz0=k0 cos θ0, ka=k0√εt − sin2 θ0, and kb=k0
√εt − εt/εz sin2 θ0. The explicit ex-
pressions for the reflection coefficients in (4.65) and (5.18) have not been found in the
literature.
A simple check of (5.2), (5.3), and (5.5) reveals that when the filling fraction approaches
zero (0), the permittivities εt and εz approach the permittivity of a homogeneous back-
ground, εa. When the filling fraction approaches one (1), the permittivities εt and εz
approach the permittivity of the materials blocks or inclusions, εb. Consequently, the so-
lutions for the horizontal and vertical reflection coefficients in (5.18) reduce to the exact
horizontal and vertical reflection coefficients at the surface of the grounded dielectric layer
of thickness t and relative permittivity εr found in (4.65a) and (4.65b).
As the filling fraction varies from one (1) to zero (0) for electrically small cells, the phase
angle of the horizontal and vertical reflection coefficients determined using the method of
moments solution should vary between the exact value obtained from a grounded dielectric
layer and the exact value obtained from a ground plane alone. The phase angles of the
horizontal and vertical reflection coefficients for grounded substrates having filling fractions
between these bounds are calculated in columns two through five of Table 5.2 using the
IE/MoM solution employing polarization currents. The values listed in columns two and
three are calculated using 192 unknown subsectional bases (Nx=Ny=Nz=4 for each of the
three polarizations) and in columns four and five using 288 unknown subsectional bases
(Nx=Ny=4 and Nz=6 for each polarization). As was mentioned at the end of Chapter 4,
fewer unknowns per wavelength are required to model the transverse variation to obtain
a given accuracy because the fields are more smoothly varying in the transverse direction.
The phase angles for the equivalent grounded uniaxial medium are calculated using the
analytical solutions in (5.18) and are listed in columns six and seven. The longitudinal
and transverse components of the permittivity tensor calculated from (5.2) and (5.6) are
listed in columns eight and nine of Table 5.2. Both solutions are obtained for a grounded
dielectric layer of relative permittivity εr=2.56 with thickness t=0.15λd, incident angles
φ0=45 and θ0=45, and unit cell size of a=0.25λ0 at z0=20t. Again, the solutions were
121
Table 5.2: Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of rel-ative permittivity εr=2.56 and dielectric thickness t=0.15λd as a function offilling fraction b/a for 192 unknowns and 288 unknowns for Np=Nq=61, φ0=45,θ0=45, and a=0.25λ0
IE/MoM IE/MoM Uniaxial
(Nx=Ny=Nz=4) (Nx=Ny=4, Nz=6)
b/a Rh Rv Rh Rv Rh Rv εz εt
1.00 -67.00 87.30 -66.20 88.67 -64.99 89.52 2.56 2.56
0.90 -63.42 95.15 -62.87 95.20 -61.43 95.25 2.26 2.10
0.75 -59.90 103.77 -59.60 103.30 -58.40 102.05 1.88 1.65
0.50 -56.56 115.48 -56.48 114.79 -55.95 112.16 1.39 1.25
0.25 -55.01 123.33 -55.00 123.83 -54.90 121.27 1.10 1.06
0.00 -54.59 125.41 -54.59 125.41 -54.59 125.41 1.00 1.00
determined using at most 3721 Floquet modes (Np=Nq=61). However, the error criterion
given in (4.76) of εc < 0.001 was often satisfied by far fewer Floquet contributions. For the
data shown in Tables 5.2–5.4, the magnitudes of Rh and Rv are close to the expected value
of 1. Although the IE/MoM solution procedure is valid for any number of subsectional
bases, the effort to evaluate the impedance matrix elements for unit cell spacings larger
than a=0.25λ0 can become so large as to be computationally difficult. Fortunately, the
accuracy of the emulated uniaxial model increases with decreasing unit cell size (electrically
small unit cells). The accuracy of the magnitude and phase of the horizontal and vertical
reflection coefficients determined from the IE/MoM solutions suggests the model for the
equivalent uniaxial medium is valid.
In Table 5.3, the computation of the horizontal and vertical reflection coefficients is
carried out for the same parameter set as in the previous table but for an asymmetric
incident angle (φ0=60 and θ0=30) and is seen to again compare favorably with the solution
obtained for the uniaxial model.
In Table 5.4, the computation of the horizontal and vertical reflection coefficients is car-
ried out for the same parameters as Table 5.2 but for a dielectric constant of εr=6.15 and
is seen to again compare favorably with the solution obtained for the uniaxial model. How-
ever, the solution obtained using only 192 unknowns (Nx=Ny=Nz=4 for each of the three
polarizations) is less accurate than the case tabulated in Table 5.2. This is to be expected
122
Table 5.3: Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of relativepermittivity εr=2.56 and dielectric thickness t=0.15λd as a function of fillingfraction b/a for 188 unknowns, Np=Nq=61, φ0=60, θ0=30, and a=0.25λ0
IE/MoM Uniaxial
b/a Rh Rv Rh Rv εz εt
1.00 76.75 -113.02 78.49 -111.78 2.56 2.56
0.90 80.58 -107.70 82.70 -106.71 2.26 2.10
0.75 84.62 -101.67 86.30 -101.56 1.88 1.65
0.50 88.52 -94.52 89.24 -95.48 1.39 1.25
0.25 90.36 -90.32 90.50 -91.00 1.10 1.06
0.00 90.87 -89.13 90.87 -89.13 1.00 1.00
since the electrical length in the dielectric material increases with increasing permittivity.
The solution obtained using 288 unknowns (Nx=Ny=4 and Nz=6 for each polarization) is
more accurate and suggests the model for the equivalent uniaxial medium is valid.
Table 5.4: Phase angle for Rh and Rv at z0=20t for a grounded dielectric layer of rel-ative permittivity εr=6.15 and dielectric thickness t=0.15λd as a function offilling fraction b/a for 192 unknowns and 288 unknowns for Np=Nq=61, φ0=45,θ0=45, and a=0.25λ0
IE/MoM IE/MoM Uniaxial
(Nx=Ny=Nz=4) (Nx=Ny=4, Nz=6)
b/a Rh Rv Rh Rv Rh Rv εz εt
1.00 -86.95 63.27 -88.05 64.12 -86.45 65.42 6.15 6.15
0.90 -85.45 71.96 -84.17 72.52 -80.74 74.19 5.17 3.80
0.75 -81.19 81.21 -80.50 80.84 -78.05 79.83 3.90 2.37
0.50 -77.51 93.40 -77.42 92.43 -76.57 86.96 2.29 1.46
0.25 -76.15 101.71 -76.15 101.27 -76.05 96.76 1.32 1.11
0.00 -75.91 104.09 -75.91 1104.09 -75.91 104.09 1.00 1.00
123
5.4 Rectangular Patch Antenna on a Positive Uniaxial Sub-
strate
Some general observations concerning patch antenna performance are in order before
the effect of anisotropy is addressed. The bandwidth and efficiency of a patch are increased
by increasing the thickness of the substrate or decreasing the relative dielectric constant.
This statement is physically reconciled by viewing the patch antenna as a resonant cavity
with two radiating slots at the end of the patch. As the radiating cavity is more loosely
bound either by increasing the substrate thickness or by decreasing the permittivity, the Q
of the resonator lessens; thus, the bandwidth widens. However, as was mentioned in the in-
troduction, integrating antenna structures on high permittivity substrates is advantageous
for circuit integration and for minimizing circuit/antenna size. The resulting consequence
of using the higher permittivity substrates is that the potential radiating energy is confined
even more tightly, narrowing the bandwidth. Higher permittivity substrates also necessi-
tates additionally thinning the substrate to reduce the formation of unwanted surface waves,
further reducing the bandwidth and efficiency.
Additionally, many of the practical substrates in use today have some amount of (uni-
axial) anisotropy. Anisotropy occurs naturally in some materials, whereas in others, it is
artificially produced in the manufacturing process. The characterization of the effect of
anisotropy on resonant length (frequency), bandwidth, and other design parameters is of
importance to the microwave community. The primary effect of uniaxial anisotropy on
patch antennas is the change in resonant length (frequency). This is significant because of
the narrow bandwidth of the patch itself. The relatively large shift in resonant frequency
produced in many of the modern substrates may actually force a rectangular patch designed
to operate at a specific frequency to radiate outside of the antenna bandwidth [74].
In Figure 5.5 is depicted an example design where the patch length L is 2.29 cm, the
patch width W is 1.9 cm, and the substrate thickness d is 0.159 cm. A coaxial probe is used
to feed the patch. It is necessary to match the probe impedance to the input impedance of
the patch. The input impedance ZA at the edge of a radiating patch is approximated [110]
as
ZA = 90ε2r
εr − 1
(L
W
)2
(Ω) (5.19)
where εr is the permittivity of the homogeneous substrate. For the example patch of
Figure 5.5, ZA=163 Ω. In order to properly match the coax probe to the patch, the probe
124
L
W
xy
z
d
(x ,y )p p
Figure 5.5: Geometry for a rectangular patch on a grounded dielectric substrate
is inset a distance ∆xp from the patch edge. The input resistance of (5.19) is reduced by
the factor cos2 (π∆xp/L) as the distance ∆xp from the edge is increased [37]. To correctly
match a 50 Ω coaxial probe to the example patch, the inset distance should be 0.40 cm
from the radiating edge.
For a rectangular patch on a homogeneous substrate with relative dielectric constant εr,
the lowest resonant frequency can be accurately predicted from [7] as
fr =c
2 (L+ 2∆L)√εeff
(5.20a)
where
∆L = 0.412d(εeff + 0.3)
(εeff − 0.256)(W/d+ 0.264)(W/d+ 0.8)
(5.20b)
and
εeff =εr + 1
2+εr − 1
21√
1 + 12d/W, for W/h >> 1. (5.20c)
Thus, for the patch design of Figure 5.5, the effective dielectric constant is εeff=2.13 and
the resonant frequency fr is predicted to be 4.194 GHz. A simulation of the patch using
Ansoft HFSS finds the lowest resonant frequency to be 4.125 GHz. This is close to the
value of 4.123 GHz calculated in [14] using a spectral domain MoM approach and also to
the measured value of 4.014 GHz given in [7]. The bandwidth BW of the resonant patch
can be approximated from [110] as
BW = 3.77εr − 1ε2r
W
L
d
λ. (5.21)
For the example design shown in Figure 5.5, the bandwidth is 1.85%.
125
When the anisotropic ratio is decreased to 0.5 (εt=2.32, εz=4.64), the resonant fre-
quency found from the finite element simulation shifts to the lower frequency of 3.015 GHz.
This value is consistent with the value of 3.032 GHz given in [14] for the same AR shift.
However, for the same AR=0.5 but εz=2.32, the resonant frequency shifts to 4.200 GHz
and the fractional change is on the order of the bandwidth of the antenna (∆f/fr=1.82).
This value is also consistent with with the value of 4.175 GHz given in [14] for the same
configuration. However, if the AR remains fixed and εz is increased to 4.64, the fractional
change in resonant frequency is significantly larger than the bandwidth. The dependence
of the resonant frequency fr and bandwidth ∆f/fr on the anisotropy for the rectangular
patch design is presented in Table 5.5.
Table 5.5: Dependence of resonant frequency on substrate permittivity
εt εz AR fr (GHz) ∆f/fr
2.32 2.32 1 4.125 0
1.16 2.32 0.5 4.200 1.82
2.32 4.64 0.5 3.015 26.9
Unlike (5.20), no general expression has been found to predict patch behavior on a
uniaxial substrate. However, it is well-known that the effect of anisotropy increases with
substrate thickness. While a significant amount of energy is coupled into the z -component
of the electric field of the TM modes for electrically small substrates, very little energy is
directed into the TE modes. However, as the substrate thickness increases, more energy
is coupled into the transverse component of the electric field that comprise the TE modes.
Consequently, the dependence on the transverse permittivity is increased and the effect
of the anisotropy more pronounced [74, 123]. Hence, for electrically thin substrates, the
dependence of design parameters such as resonant length and bandwidth should be strongly
dependent only on εz.
If, for the previous example where the anisotropic ratio has been decreased to 0.5 and
εz=4.64, the relative dielectric constant in (5.20) is replaced with the corresponding εz, the
predicted resonant frequency of the patch is 3.042 GHz. This is reasonably close to the
value determined from the simulation and to the value of 3.032 GHz given in [14]. This
observation is also consistent with results presented in [14] that the fractional change in
the resonant frequency is strongly dependent on εz. By modeling the uniaxial material as
a doubly periodic dielectric layer, the understanding of this dependence can be shown even
126
more clearly.
Consider a rectangular patch integrated on a uniaxial substrate. The uniaxial substrate
can be modelled as a homogeneous dielectric layer with periodic inclusions in two directions.
Because the axial and transverse components of the permittivity tensor of the uniaxial sub-
strate are known, the homogeneous permittivity and filling fraction of the doubly periodic
substrate can be found through (5.5) and (5.2). For an electrically thin substrate, the
effect of removing material from underneath the patch is a volumetric “averaging” of the
dielectric constant. This assumption is intuitively appealing and is probably valid for thin
substrates, but may not hold as the thickness of the substrate increases (and consequently
the dependence on the transverse permittivity).
As was pointed out earlier, the equivalent axial component of the permittivity tensor of
the emulated periodic substrate is effectively the volumetric average of the two phases. For
example, sapphire which has an AR of 0.81 (εt=9.4, εz=11.6) is modelled in the HFSS ma-
terial library as a homogeneous substrate with relative dielectric constant εr=10. However,
sapphire can be modeled by a medium consisting of periodic air inclusions in a dielec-
tric background with dielectric constant of 16.0 and filling fraction equal to 0.3 as seen in
Figure 5.4(b). If one simply takes the volume averages of the two phases as the effective
permittivity εeff of the equivalent substrate, one finds that εeff=11.6 which is very close the
value of the axial component of the permittivity tensor. This effective permittivity is not
to be confused with the equivalent homogeneous permittivity that can be determined from
(5.20). Clearly, modeling a sapphire substrate as a homogeneous substrate with an “effec-
tive” permittivity would yield a very different resonant frequency than if the uniaxial nature
of the material is taken into account. In fact, the lowest resonant frequency of the patch
shown in Figure 5.5 when integrated on a substrate of permittivity εr=10 yields a predicted
resonant frequency from (5.20) of 2.173 GHz. However, if the true uniaxial nature is taken
into account, the resonant frequency should be 2.022 GHz. Since the resonant frequency
and electrical length of the patch are directly related, one can interpret the effect of the
increased axial component as an effective lengthening of the patch. This interpretation is
confirmed when one views the patterns produced by the patch antenna.
Anisotropic effects shape the radiation pattern of the patch and thus in an array config-
uration, the coupling to other elements. For the example depicted in Figure 5.5, the E - and
H -plane patterns are determined for both homogeneous (εt=2.32, εz=2.32) and anisotropic
(εt=2.32, εz=4.64) substrates. The H -plane patterns shown in Figure 5.6(b) are relatively
independent of the AR and εz. However, the E -plane patterns shown in Figure 5.6(a)
127
Directivity (dB
)
-20
-10
00
30
60
90et=2.32, ez=2.32
εt=1.16, εz=2.32
εt=2.32, εz=4.64
(a) E -plane
Directivity (dB
)
-20
-10
0
0
30
60
90
εt=2.32, εz=2.32
εt=1.16, εz=2.32
εt=2.32, εz=4.64
(b) H -plane
Figure 5.6: Directivity patterns for reference patch with AR=1 and AR=0.5
show a strong dependence on εz. This is intuitively correct if one assumes that for thin
substrates, the dependence is on εz alone. Thus, if the true axial component is ignored for
substrates with positive anisotropic ratios, the beamwidth will be wider than designed and
consequently the mutual coupling in an array environment may affect the performance of
such an antenna. The observation that the E -plane beamwidth is affected by the anisotropy
while the H -plane beamwidth is not affected by it is consistent with approximations for the
beamwidths given in [11]. The E -plane and H -plane beamwidths can be approximated by
[11]
ΘE 2 cos−1
√7.03λ2
0
4(3L2e + d2)π2
(5.22a)
ΘH 2 cos−1
√1
2 + k0W(5.22b)
where Le is the effective length of the patch. As the effective length of the patch is increased,
the E -plane beamwidth decreases. The H -plane beamwidth is independent of Le.
5.5 Conclusions
Because uniaxial substrates are often expensive to manufacture and have limited flexibil-
ity for design, uniaxial substrates can be emulated (and easily fabricated) by incorporating
128
periodic inclusions in an otherwise homogeneous substrate. The doubly periodic structure
can be easily constructed using simple milling or etching techniques from simple inexpen-
sive, homogeneous substrates. This is significant because some common uniaxial materials
such as sapphire that are expensive to grow can be “artificially” replicated easily and in-
expensively. Additionally, the artificial nature of the periodic uniaxial substrate permits
the creative design of new substrates with the expanded freedom of anisotropic ratio, back-
ground permittivity, and/or fabrication technique. The primary effect of anisotropy on
rectangular patch antennas is the change in its resonant length (frequency). This is signif-
icant because of the narrow bandwidth of the patch. The relatively large shift in resonant
frequency produced in many of the modern substrates may actually force a rectangular patch
designed to operate at a specific frequency to radiate outside of the antenna bandwidth.
In addition to providing freedom of design, modeling uniaxial materials using the doubly
periodic model provides an insightful understanding of the dependence of the performance
of the substrate on its uniaxial nature.
129
CHAPTER 6
Recommendations For Future Work
One of the more interesting physical features that can be altered in the solution of
the one- and two-dimensional periodic media is the specification of an arbitrarily shaped
dielectric. Both the IE/MoM solution and plane wave expansion method can be adapted to
provide for optimized dielectric functions. Two potential applications for effective medium
theory that promise to be of value are the extension of EMT to off-axis propagation in
two-dimensional lattices and to out-of-plane propagation in two-dimensional lattices. A
number of new applications can be developed using the solution technique implemented in
Chapter 4 and 5 including new frequency selective volumes, incorporating noncommensurate
periodicities for each layer, incorporating material implants within each layer of differing
relative permittivity (dielectric and/or metallic loading), and the extension of the solution
to large planar antenna elements. The formulation also has the flexibility of providing that
each layer have different element shape and/or permittivity.
6.1 Arbitrarily Shaped Dielectric Function
By optimizing the dielectric function, improvements in design and performance can be
achieved for applications at both optical and microwave frequencies. The optimization is
accomplished through the use of genetic algorithms, neural networks, and other “smart”
algorithms. The design of a unique dielectric function is achieved by allowing the opti-
mizing algorithm or morphing program to vary the different parameters, such as unit cell
size, filling fraction, element shape, and/or relative permittivity, until a desired response
is obtained. Using the IE/MoM solution, unique configurations can be formed by selec-
tively removing individual polarization currents from the solution, commonly referred to
in antenna engineering as thinning. This is accomplished in a similar fashion to thinning
130
the number of elements in a phased array antenna. For the plane wave expansion method,
the optimization is accomplished by computing the fast Fourier transform (FFT) of the
discretized dielectric function. Each time the dielectric function is changed, a new FFT is
performed to determine the new Fourier coefficients.
Figure 6.1: Optimized frequency selective layer (volume)
New materials with unique spectral characteristics are constantly being developed and
can be incorporated into existing designs that in turn can be optimized for performance.
An example of a structure that might be produced by an optimizing algorithm is seen in
Figure 6.1, where specific individual blocks of material have been removed from a large
unit cell. The optimized structure might have application as a frequency selective layer
(volume) for use in radome applications. Additionally, lattices of dielectric elements with
differing dielectric constants can be included to provide additional freedom in the design.
In the IE/MoM solution, the polarization currents incorporate the permittivity change.
In the plane wave expansion method, a new dielectric function must be developed. As
illustrated in Figure 6.2, electrical loading of the layer is realized by incorporating two
dielectric constant 1
dielectric constant 2
Figure 6.2: Dielectric and/or metallic loading
different dielectrics or perhaps including metallic regions in the design. Currently, an entire
131
class of PBG materials, metallic photonic band-gap (MPBG) materials composed of metallic
lattice elements, are being investigated [62].
Another application where optimizing the dielectric function might be useful is supercell
theory. A supercell is a unique geometrical configuration comprised of two or more unit
cells, each with same structure. An example of a supercell can be seen in Figure 6.3 where
the individual unit cell of Figure 6.1 has been periodically repeated. In order that the
spectral response of the supercell layer be distinct from the response of a uniform layer of
effective permittivity, the spacing between the individual elements must be on the order of
a quarter of a wavelength. Consequently, applications that use the supercell design have
the disadvantage of being electrically and, depending on the frequency, physically large.
Although the response of the supercell might be excellent, the electrical and physical size
of the resulting structure might by impractical.
Figure 6.3: Frequency selective supercell
6.2 Extensions of Effective Medium Theory
To simplify the computation of the band structure for two-dimensionally periodic media,
effective medium theory was applied in Chapter 3 to reduce the two-dimensional periodic
structure to a one-dimensional equivalent structure. Using EMT, each periodic row is re-
placed by a thin homogeneous layer of effective permittivity [50] which is determined solely
as a function of the geometrical and electrical parameters of the lattice. Unfortunately, tra-
ditional EMT is restricted to near normal incidence precluding its use for many applications
of interest. Consider the two-dimensional lattice with dielectric constant ε2 immersed in a
background with dielectric constant ε1 illustrated in cross-section in Figure 6.4. Assume a
plane wave is incident in a direction along a ray direction defined by an angle φ with respect
132
x
y
ε eff
ε eff
ε eff
ε eff
ε2
ε1
E(H)φ
Figure 6.4: Higher-order effective medium theory
to the x axis (φ = 0). An extension of EMT to off-axis propagation would allow EMT to
be useful in some planar antenna applications. Another effective EMT extension would be
to model the hybrid modes that are formed for off-plane propagation in two-dimensional
lattices.
6.3 Noncommensurate Periodicities
a
b
Figure 6.5: Noncommensurate periodicities
The solution of scattering from a multilayered medium is often accomplished using a
generalized scattering matrix (GSM) technique [136] which relates the modes that propagate
133
in one layer to the corresponding modes in adjacent layers. The solution of scattering from
a multilayered medium where each layer has a unique period has only recently be solved and
then only for commensurate periodicities [12]. An example of a multilayered medium where
each layer has a unique period is illustrated in Figure 6.5. The solution found in [12] relates
the local Floquet modes within each layer to the global Floquet modes in the structure.
The solution is straightforward but relies on a tremendous amount of bookkeepping. A new
solution to this problem would be of tremendous value.
6.4 Extensive Parametric Study of Antenna Performance Near
a Periodic Structure
An extensive parametric study of the factors that impact the performance of an antenna
integrated on a periodic surface would be valuable to the microwave community. The
salient properties of the antenna, such as its input impedance, bandwidth, beamwidth,
directivity, gain, efficiency, polarization, and mutual coupling in an array environment, are
all determined by the antenna’s electrical size, physical configuration, and the environment
in which it is located. For planar resonant patches alone, the designer has a multiplicity
of factors that must be considered. These include, but are not limited to, the material
properties of the substrate such as dielectric constant, loss tangent, anisotropy, and the ease
of fabrication; the physical considerations of patch shape, patch size, and system integration;
and for realizable antennas, the practical considerations of feed location, feeding technique,
and design cost.
Although some attention has been directed at patch antenna performance on uniaxial
substrates, it has been limited to idealized cases and simple observations. The need exists
for a complete treatment of antennas located near periodic surfaces and structures. This
can only be accomplished if extensive parametric studies are performed to determine the re-
sulting effects. For example, planar antennas mounted on substrates of significant thickness
(often used to increase bandwidth) radiate most of their energy into the substrate creating
surface waves that propagate in specific directions. The reduction or elimination of these
surface waves in planar antenna application has been the Holy Grail of microwave antenna
engineers for many years. If an appropriate design for a periodic substrate that supports
the planar antenna can be found that reduces or eliminates the formation of these waves,
significant improvements in performance will be achieved.
As the need for very wideband, omni-directional antennas for use in mobile communi-
134
cation networks grows at an increasing rate, so does the need to develop new fabrication
technologies, advanced materials engineering, and novel antenna architectures to meet the
challenge. Tomorrow’s RF engineer has the task of developing new and useful ideas in
addition to developing previously found materials and concepts to their fullest potential.
135
APPENDICES
136
APPENDIX A
Transmission Line Formulation for the One-Dimensional
Periodic Array of Dielectric Slabs
Transmission line theory allows one to formulate a simple solution for the propagation
of transverse electromagnetic waves (TEM) through periodic dielectric slabs. The dielectric
slabs and free-space regions are modeled by short sections of transmission line as seen in
Figure A.1(a) where b is the length of the shorter section BB′ with characteristic impedance
Zd and propagation constant γd=jβd, a is the length of the entire transmission line section
AA′, Z0 is the characteristic impedance of the free-space transmission line sections, jβ0 is
the propagation constant of the free-space sections, and d = (b− a)/2. Due to the periodic
nature of the structure, the propagation constant through the structure can be determined
by analyzing one unit cell. For waves propagating through the structure, the voltage VA′
and current IA′ at the A′ plane is required to be the same, except for a propagation factor,
as the value of VA and IA at the A plane, or
VA′ = VAe−γa and IA′ = I−γaA (A.1)
where γ = α + jβ is the propagation constant of the line and a is the length of the line.
The input impedance at the A and A′ planes is defined as
Zp =VA
IA=
VA′
IA′. (A.2)
The equivalent transmission line model is shown in Figure A.1(b).
137
Z
+
_0 Z0Zd
A A’B B’I A I A’
+
_
VA VA’
b
a
(a) Loaded transmission line circuit
Z
+
_p
A A’I A I A’
+
_
VA VA’
a
(b) Equivalent transmission line circuit
Figure A.1: Loaded and equivalent transmission line circuits
The ABCD matrix for the transmission line in Figure A.1(a) is
A = cosh γdb cos 2β0d+ j(Z2d + Z
20
2ZdZ0
)sinh γdb sin 2β0d (A.3)
B = jZ0 cosh γdb sin 2β0d+ sinh γd
(Z2d
Z0cos2 β0d− Z2
0
Zdsin2 β0d
)(A.4)
C =j
Z0cosh γdb sin 2β0d+ sinh γd
(Z2
0
Zdcos2 β0d− Zd
Z20
sin2 β0d
)(A.5)
D = cosh γdb cos 2β0d+ j(Z2d + Z
20
2ZdZ0
)sinh γdb sin 2β0d. (A.6)
The ABCD matrix for the equivalent transmission line in Figure A.1(b) is
A = cosh γa (A.7)
B = Zp sinh γa (A.8)
C =1Zp
sinh γa (A.9)
D = cosh γa (A.10)
where Zp is the effective input impedance of the equivalent section and γ is the propagation
constant of the line.
138
The propagation constant γ of the equivalent circuit, and thus, for the periodic struc-
ture, is obtained easily from the two ABCD matrices to be the solution of the following
transcendental equation
cosh γa = cosh γdb cos 2β0d+ j(Z2d + Z
20
2ZdZ0
)sinh γdb sin 2β0d (A.11)
which can be solved using a numerical solver such as MATLAB. The input impedance of
the line can be found by equating the ratio of (A.4) and (A.5) and the ratio of (A.8) and
(A.9)
Zp = Z0
jZ0 cosh γdb sin 2β0d+ sinh γdb
(Z2
dZ0
cos2 β0d− Z20
Zdsin2 β0d
)jZ0
cosh γdb sin 2β0d+ sinh γdb(Z2
0Zd
cos2 β0d− Z2d
Z0sin2 β0d
)
1/2
. (A.12)
The frequencies where the right hand side of (A.11) returns a value greater than unity
and γ is purely real (γ = α) define the stopband of the structure. At frequencies where
cosh γa < 1, the propagation constant is purely imaginary (γ = jβ). These frequencies
define the passband of the structure. The results obtained from these equations yield
solutions that agree with the results obtained using the method moments solution and the
solution of the exact eigenvalue equation. Additionally, the solutions obtained in (A.11)
and (A.12) are identical to (9-153) and (9-154) in [78] with the exception that the solutions
contained herein are more general, being derived for a short section of reactively loaded line
of length b, as opposed to an inductively loaded case as in [78].
Other useful forms, analogies, and derivations of the exact dispersion relation for one-
dimensionally periodic dielectric slabs can be found in [16, 55, 107, 26].
139
APPENDIX B
Bravais Lattices and the Brillouin Zone
An important concept in propagation through two- and three-dimensional periodic me-
dia are direct and reciprocal lattices. For two-dimensional periodic structures, the direct
lattice must belong to one of the five two-dimensional Bravais (or space) lattices illus-
trated in Figure B.3 on page 143. Three-dimensional lattices (crystals) have 32 unique
space lattices (with three dimensions of freedom) that can be formed. A discussion of
three-dimensional crystalline structures is beyond the scope of this work and the reader is
referred to an introductory text on solid state physics such as Kittel [44].
Direct lattice
The direct lattice in Figure B.1(a) is defined by the length of the two primitive lattice
vectors a and b and the angle γ between the vectors in the plane perpendicular to a × b.
The unit cell of Figure B.1 is completed by the dashed lines, has an area of |a||b| sin γ, and
γb
a
(a) Direct lattice
b*
a*
(b) Reciprocal lattice
Figure B.1: Direct lattice defined using the primitive lattice vectors a and b and the re-ciprocal lattice defined using the primitive reciprocal lattice vectors a∗ andb∗
140
is invariant under the translation ma+nb for any integers m,n. For the two-dimensional
structures implemented in this work, the primitive lattice vectors represent the physical
distance between the periodic elements (dielectric rods or air columns).
Reciprocal lattice
In many instances the solution of a problem defined on a periodic lattice is more con-
venient to implement in Fourier space. Thus, it is convenient to define a reciprocal lattice
space. The reciprocal lattice in Figure B.1(b) is defined by two primitive reciprocal lattice
vectors a∗ and b∗. The construction of a∗ and b∗ is simple — the only requirement being
that a ·a∗=b ·b∗=2π and a ·b∗=a∗ ·b=0. As a consequence of using plane waves to expand
the periodic function, the normalization is set equal to 2π.
Example 1. Square lattice of Figure B.3(b) with spacing d
a = −d y b = d x
a∗ = −2πd
y b∗ =2πd
x
Example 2. Hexagonal lattice of Figure B.3(c) with spacing d
a = −d2
x− d√3
2y b = d x
a∗ = − 4πd√3
y b∗ =2πd
x− 2πd√3
y
Clearly, for both examples, a · a∗=b · b∗=2π and a · b∗=a∗ · b=0.
Brillouin Zone
Periodic structures, particularly those with two- and three-dimensional periodicity, are
by definition highly symmetric. Consequently, propagation through such media necessar-
ily contains many redundant propagation vectors. The unique vectors can be found and
categorized systematically using the concept of the Brillouin zone (BZ) [44] shown for a
square regular lattice in Figure B.2. For the square lattice, all of the possible directions of
propagation are grouped into eight regions with specific symmetry.2 However, symmetry
considerations reduce the number of unique regions of propagation to one. The colored re-
gion in the figure is called the irreducible Brillouin zone and represents the smallest region
where propagation within the lattice is unique. This region is further defined by symmetric2For the hexagonal lattice of Example 2, the Brillouin zone contains six groupings.
141
ΓX
M
k=k x
k=π/a x+k y
k=k x+k y
x
x y
y
0 π/a
π/a
0
Figure B.2: Irreducible Brillouin Zone (BZ)
points within the lattice denoted Γ, M, and X. The Γ coordinate is defined by (kx, ky)=(0, 0),
the M coordinate by (kx, ky)=(2π/a, 0), and the X coordinate by (kx, ky)=(2π/a, 2π/a). In
order to determine whether a mode is allowed to propagate, every possible vector that is
located within the irreducible Brillouin zone must be checked. Fortunately, the band struc-
ture can be determined approximately by sampling the edges of the zone along the Γ–M line
for modes of the form k = kxx+0 y, the M–X line for modes of the form k = 2π/a x+ kyy
and the Γ–X line for modes of the form k = kxx+ kyy.
142
ab
γ
(a) Oblique lattice: |a| = |b|; γ = 90
abγ
(b) Square lattice: |a| = |b|; γ = 90
ab
γ
(c) Hexagonal lattice: |a| = |b|; γ = 120
abγ
(d) Rectangular lattice: |a| = |b|;γ = 90
b
a
γ
(e) Centered rectangular lattice: |a| = |b|; γ = 90
Figure B.3: Two-dimensional Bravais (or space) lattices
143
APPENDIX C
Impedance Matrix Elements for Two-Dimensional Periodic
Structures
TMz Case
Piecewise constant expansion / Piecewise constant testing
Zmn =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.1)
δZ = − δmn∆x∆y
jk0Y0 (εr − 1). (C.2)
Piecewise linear expansion / Piecewise linear testing
Z11 =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.3)
Z1n =jk0Z0∆2
x∆2y
ac
∑p,q
sinc4(kxp∆x
2
)sinc4
(kyq∆y
2
)e−jkxp(x1−xn)e−jkyq (y1−yn)
k20 − k2
xp− k2
yq
(C.4)
Z1N =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)e−jkxp(x1−xN )e−jkyq (y1−yN )
k20 − k2
xp− k2
yq
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 (C.5)
144
Zm1 =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)e−jkxp(xm−x1)e−jkyq (ym−y1)
k20 − k2
xp− k2
yq
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 (C.6)
Zmn =jk0Z0∆2
x∆2y
ac
∑p,q
sinc4(kxp∆x
2
)sinc4
(kyq∆y
2
)e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.7)
ZmN =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)e−jkxp(xm−xN )e−jkyq (ym−yN )
k20 − k2
xp− k2
yq
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 (C.8)
ZN1 =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)e−jkxp(xN−x1)e−jkyq (yN−y1)
k20 − k2
xp− k2
yq
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 (C.9)
ZNn =jk0Z0∆2
x∆2y
ac
∑p,q
sinc4(kxp∆x
2
)sinc4
(kyq∆y
2
)e−jkxp(xN−xn)e−jkyq (yN−yn)
k20 − k2
xp− k2
yq
(C.10)
ZNN =jk0Z0∆2
x∆2y
ac
∑p,q
sinc2(kxp∆x
2
)sinc2
(kyq∆y
2
)1
k20 − k2
xp− k2
yq
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 (C.11)
δZ = − δmn∆x∆y
jk0Y0 (εr − 1). (C.12)
TEz Case
Piecewise linear expansion / Piecewise linear testing
Zxx11 =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)1
k20 − k2
xp− k2
yq
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 (C.13)
145
Zxx1n =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(x1−xn)e−jkyq (y1−yn)
k20 − k2
xp− k2
yq
(C.14)
Zxx1N =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(x1−xN )e−jkyq (y1−yN )
k20 − k2
xp− k2
yq
(C.15)
Zxxm1 =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xm−x1)e−jkyq (ym−y1)
k20 − k2
xp− k2
yq
(C.16)
Zxxmn =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
timese−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.17)
ZxxmN =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(xm−xN )e−jkyq (ym−yN )
k20 − k2
xp− k2
yq
(C.18)
ZxxN1 =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xN−x1)e−jkyq (yN−y1)
k20 − k2
xp− k2
yq
(C.19)
ZxxNn =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xN−xn)e−jkyq (yN−yn)
k20 − k2
xp− k2
yq
(C.20)
146
ZxxNN =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.21)
Zxy11 =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.22)
Zxy1n =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(x1−xn)e−jkyq (y1−yn)
k20 − k2
xp− k2
yq
(C.23)
Zxy1N =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(x1−xN )e−jkyq (y1−yN )
k20 − k2
xp− k2
yq
(C.24)
Zxym1 =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xm−x1)e−jkyq (ym−y1)
k20 − k2
xp− k2
yq
(C.25)
Zxymn =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.26)
ZxymN =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(xm−xN )e−jkyq (ym−yN )
k20 − k2
xp− k2
yq
(C.27)
147
ZxyN1 =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xN−x1)e−jkyq (yN−y1)
k20 − k2
xp− k2
yq
(C.28)
ZxyNn =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xN−xn)e−jkyq (yN−yn)
k20 − k2
xp− k2
yq
(C.29)
ZxyNN =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.30)
Zyx11 =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.31)
Zyx1n =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(x1−xn)e−jkyq (y1−yn)
k20 − k2
xp− k2
yq
(C.32)
Zyx1N =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(x1−xN )e−jkyq (y1−yN )
k20 − k2
xp− k2
yq
(C.33)
Zyxm1 =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xm−x1)e−jkyq (ym−y1)
k20 − k2
xp− k2
yq
(C.34)
148
Zyxmn =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.35)
ZyxmN =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(xm−xN )e−jkyq (ym−yN )
k20 − k2
xp− k2
yq
(C.36)
ZyxN1 =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xN−x1)e−jkyq (yN−y1)
k20 − k2
xp− k2
yq
(C.37)
ZyxNn =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xN−xn)e−jkyq (yN−yn)
k20 − k2
xp− k2
yq
(C.38)
ZyxNN =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.39)
Zyy11 =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.40)
Zyy1n =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(x1−xn)e−jkyq (y1−yn)
k20 − k2
xp− k2
yq
(C.41)
149
Zyy1N =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(x1−xN )e−jkyq (y1−yN )
k20 − k2
xp− k2
yq
(C.42)
Zyym1 =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xm−x1)e−jkyq (ym−y1)
k20 − k2
xp− k2
yq
(C.43)
Zyymn =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.44)
ZyymN =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 e−jkxp(xm−xN )e−jkyq (ym−yN )
k20 − k2
xp− k2
yq
(C.45)
ZyyN1 =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1 + jkxp∆x − ejkxp∆x
)(kxp∆x
)2(1 + jkyq∆y − ejkyq∆y
)(kyq∆y
)2 e−jkxp(xN−x1)e−jkyq (yN−y1)
k20 − k2
xp− k2
yq
(C.46)
ZyyNn =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc4
(kxp∆x
2
)sinc4
(kyq∆y
2
)
× e−jkxp(xN−xn)e−jkyq (yN−yn)
k20 − k2
xp− k2
yq
(C.47)
ZyyNN =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
×(1− jkxp∆x − e−jkxp∆x
)(kxp∆x
)2(1− jkyq∆y − e−jkyq∆y
)(kyq∆y
)2 1k20 − k2
xp− k2
yq
(C.48)
δZxx = − δmn∆x∆y
jk0Y0 (εr − 1). (C.49)
δZyy = − δmn∆x∆y
jk0Y0 (εr − 1). (C.50)
150
Piecewise constant expansion / Piecewise constant testing
Zxxmn =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
xp
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
× e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.51)
Zxymn =jk0Z0∆2
x∆2y
ac
∑p,q
(−kxpkyq
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
× e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.52)
Zyxmn = Zxymn (C.53)
Zyymn =jk0Z0∆2
x∆2y
ac
∑p,q
(1− k2
yq
k20
)sinc2
(kxp∆x
2
)sinc2
(kyq∆y
2
)
× e−jkxp(xm−xn)e−jkyq (ym−yn)
k20 − k2
xp− k2
yq
(C.54)
δZxx = − δmn∆x∆y
jk0Y0 (εr − 1). (C.55)
δZyy = − δmn∆x∆y
jk0Y0 (εr − 1). (C.56)
151
APPENDIX D
Parallel-Plate Mode Reduction In Conductor-Backed Slots
Using Periodic Dielectric Substrates
Periodic dielectric substrates offer the possibility of changing the propagation charac-
teristics of planar circuits and antennas. A number of applications for such materials can
be imagined and have been implemented including various slow-wave structures, dielectric
mirrors, resonant cavities, and frequency selective surfaces (FSS). A number of researchers
have recently begun to design electromagnetic crystal structures for use in planar antenna
applications, particularly for use as reflectors in planar dipole antenna structures. One of the
early demonstrations of the potential for these materials in the microwave and millimeter-
wave band was shown by Brown et al. in [18, 17] while investigating the radiation prop-
erties of a planar dipole on a photonic crystal substrate. Cheng et al. [21] followed by
optimizing planar dipole antennas by using PBG crystals as a perfectly reflecting planar
substrate. Kesler et al. [41] designed finite thickness slabs of two-dimensional PBG ma-
terials for use as an effective reflection plane for planar dipole antennas. The reflection
and transmission properties of two- and three-dimensional PBG materials were studied by
Sigalas et al. [95, 97] for calculating the radiation properties of planar dipole antennas.
Leung et al. [53] measured the radiation patterns of a slot antenna placed on a layer-by-
layer photonic band gap crystal. For a slot operating at a frequency in the band-gap of the
three-dimensional PBG crystal, energy which would have been radiated into the substrate
is reflected. However, at the interface between the PBG and the air, the periodicity of the
PBG is broken and a parasitic mode (surface state) can exist. These surface states decrease
the efficiency by stripping power away from the radiating element.
By fabricating a resonant slot over a reflecting back plate and filling the resulting
parallel-plate with an appropriately designed periodic dielectric substrate, noticeable en-
152
hancements in both radiation pattern and bandwidth are achieved using a significantly
lower profile than traditional designs. Measured and simulated data for conductor-backed
slots with homogeneous substrates and with periodic dielectric substrates are compared.
In order to reduce backside radiation and increase the gain of planar slot antennas,
traditional designs typically place some type of reflecting surface or cavity behind the slot.
Unfortunately, increasing the profile of the slot negates one advantage of the planar radiating
element. In addition, placing reflecting surfaces behind the slot reduces the efficiency by
creating parasitic modes and using a cavity-backed design often necessitates narrowing the
bandwidth. If the slot antenna is backed by a metal plate to increase the front-to-back
ratio, parallel-plate waveguide (PPW) modes will be excited, both decreasing the efficiency
and distorting the pattern. To completely block radiation from the backside of the slot from
propagating to the finite edges of the resulting parallel-plate cavity, the cavity can be filled
with a two-dimensional periodic dielectric substrate.
The theory and design of the periodic dielectric structure placed behind the conductor-
backed slot is developed in Section D.1. In Chapter 3, the theory of electromagnetic plane
wave propagation through simple two-dimensional periodic dielectric structures is developed
both analytically (Fourier series solution) and numerically (IE/MoM solution). The plane
wave expansion solution expands the propagating electric field as a periodic function with
a prescribed phase shift and expands the periodic dielectric rods as another periodic func-
tion with no phase shift. Consequently, the electric field and dielectric rods are expanded
in Fourier series and inserted into the wave equation. The resulting band structures can
easily be determined by solving the the resulting matrix equation for the eigenvalues of the
system. In the MoM model, the dielectric rods are replaced by equivalent (polarization)
volume currents. The total field is determined as the sum of a known incident TEM wave
propagating in a direction transverse to the cylinder axis with its electric field parallel to
the cylinder axis and a scattered field which is due to radiation by the equivalent currents
induced in the dielectric by the incident field. A nontrivial solution for the field requires
the determinant of the impedance matrix to be zero, which results in a characteristic equa-
tion. The eigenvalues (propagation constants) are obtained from the roots of this equation.
The folded slot antenna design and fabrication are outlined in Section D.2. Observations
concerning the performance of the periodic dielectric structure and its use for backing slot
antennas is detailed in Section D.3. General observations about the usefulness and possible
applications for these special materials are found in the conclusions.
153
D.1 Electromagnetic Band-Gap Theory and Design
If the reflecting plate shown in Figure D.1 is located near enough to the slot such that the
reflecting plate
electromagneticband-gap (EBG) substrate
coaxially-fed folded slot
Figure D.1: Conductor-backed folded slot with periodic dielectric substrate
operating frequency is below the cutoff of the first TE/TM mode, only the dominant TEM
mode (TM0) with zero cutoff frequency will propagate. A single TEM mode is produced
by keeping the separation distance between the plates less than half the guide wavelength.
Because the propagation characteristics of the TEM mode of the resulting waveguide are
very similar to the propagation characteristics of a uniform plane wave propagating normal
to the axis of a infinite height dielectric structure, the substrate can be modeled as a
doubly-periodic array of infinite dielectric rods in free space. The rods are imaged infinitely
in height by using a PEC (perfect electric conductor) boundary and infinitely in width by
using an PMC (perfect magnetic conductor) boundary located one-half unit cell away from
the center of the rods (Figure D.2).
D.1.1 Substrate Design
For a simple two-dimensional periodic array of dielectric cylinders, the parameter com-
binations of unit cell size, element size, element shape (square/circle), lattice class (square,
triangular, hexagonal), and dielectric contrast between the insert and the background ma-
terial are the significant contributors to the resulting band structure.
154
input port
output port
PMC
PMC
output port
PEC
PEC
Figure D.2: Simulation unit cell
Lattice shape
A square lattice of dielectric columns is chosen to both simplify the fabrication and
because it produces a larger gap for relatively lower frequencies (for parallel (TM) polar-
ization) than do traditional hexagonal or honeycomb lattices [116, 63]. A brief discussion
of other two-dimensional lattice structures can be found in Appendix B. For dielectric rod
spacings of fractions of a wavelength, the primary TM band-gap of interest is the lowest
band. The lower the lowest frequency band, the more applicable the periodic dielectric
structure is for compact circuit applications.
Element shape
Although early two-dimensional research focused on using circular rods, square rods of
commensurate size are chosen to simplify the formulation and fabrication. Band structures
calculated in Chapter 3 justify replacing the traditional circular rods by square rods in
this application. The geometries of the circular and square dielectric elements are shown
in Figure D.3. Using circular rods of diameter b=4.8 mm and square rods of edge length
b=4.8 mm produces a similar band structure (first gap) for the combination of unit cell size
a=1.2 cm and relative dielectric constant εr = 10.2. In Figure D.4, one can clearly see that
both the circular and square rods have relative large first gap-to-midgap ratios of 0.176 and
0.167, respectively. Consequently, small errors in the fabrication of the substrate can be
155
a
b
(a)
a
b
(b)
Figure D.3: Unit cell dimensions for (a) square dielectric rods and (b) circular dielectricrods where for the square rod, a is the unit cell size, b is the element edgelength, and for the circular rod, a is the unit cell size and b is the diameter
Normalized insert size (x/a)
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7SquareCircularSubstrate design
Figure D.4: Band-gap plot as a function normalized insert size b/a
absorbed with little or no consequence. The final design of the periodic structure is shown
as a circle in the center of the gaps and produces an omni-directional stopband between 7.5
and 10 GHz. The full band structure of a square lattice of dielectric columns as a function
of b/a is seen in Figure D.5 for circular or square rods.
Design validation for finite lattice
In parallel to the the method of moments solution and plane wave (Fourier series –
eigenvalue) expansion solution, finite element simulations have been implemented to cor-
156
0.00 3.14 6.28 9.42
Nor
mal
ized
fre
quen
cy (
fa/c
)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
SquareCircular
Γ βx (βy=0) X βy (βx=π/a) M βx=βy Γ
Normalized propagation constant ( βa)
Figure D.5: Full band structure of a square lattice of dielectric columns as a function ofnormalized insert size b/a
roborate the band structure of a realizable (finite) structure. In order to achieve a useful
stopband, a large (ideally, infinite) number of periods are required. However, significant
attenuations can still be achieved using only a small number of periods by designing an ap-
propriate periodic structure. The periodically dielectric-loaded parallel-plate is simulated
using the Agilent EEsof EDA High-Frequency Structure Simulator (HFSS). For a square
lattice of five periods, the simulations take between four seconds for the Γ–X direction of
the irreducible Brillouin zone1 (Figure D.6) and eight seconds for the Γ–M direction per
frequency point on a 400 MHz Pentium II PC. Because the speed of the simulation is so
Γ(k =0,k =0)
X(k =π/a,k =0)
M(k =π/a,k =π/a)x
x
xy
y
y
Figure D.6: Irreducible Brillouin Zone (BZ)
1The concept of the irreducible Brillouin zone for two-dimensional periodic structures is developed furtherin Appendix B.
157
rapid, these structures can be repetitively designed and validated quickly.
In the finite structure, the center frequency of the first stopband for the square rods
is 8.5 GHz (Γ–X direction, b/a=0.4, εr=10.2). As expected, the center frequency of the
first stopband for the circular rods is 9.0 GHz (Γ–X direction, b/a=0.4, εr=10.2), about 5%
higher than the case using square elements. However, the 10 dB bandwidth of the stopband
is 4.0 GHz for square dielectric elements and 4.7 GHz for circular dielectric elements. The
significant increase in the gap-to-midgap ratio (bandwidth) for the finite structure can be
attributed to the certain widening of the stopband – a necessary consequence of using a finite
number of elements – and to the requirement of a finite attenuation inside the stopband. At
the center frequency of the band-gap, omni-directional attenuations of at least 20 dB and
upwards of 45 dB attenuation in specific directions can be obtained using only 3–4 periods
(Figure D.7). This is compared to the ideally infinite attenuations within the stopband of
an unrealizable infinite structure.
Transmission (dB)
-20-15-10-5056789101112
Transmission (dB)
-20 -15 -10 -5 0
G MX
Fre
quen
cy (
GH
z)
1234
1234
56789101112
Fre
quen
cy (
GH
z)
Figure D.7: Simulated (HFSS) transmission spectra for Γ–X (left) and Γ–M (right) direc-tions and calculated (MoM) band diagram (center) of a two-dimensional EBGwith a=1.2 cm, b=4.8 mm, and εr=10.2 for TM polarization.
D.2 Slot Antenna Design and Fabrication
A conductor-backed folded slot with a homogeneous substrate was designed initially to
provide an acceptable match when center-fed. Preliminary work using this λg/2 reference
slot revealed the difficulty in matching the high input impedance of the slot to the coaxial
feed. Consequently, the single λg/2 slot was replaced with a folded λg slot. Not only is the
folded λg slot easier to match to the input impedance of the coaxial feed than the λg/2 slot
158
but it also has increased bandwidth.
D.2.1 Reference Slot
A half-wave reference folded slot of length 14.3 mm and width of 2.0 mm shown in
Figure D.8, where slot dimensions in parentheses refer particularly to the reference slot
offset feed
18.0 mm(14.3 mm)
2.0 mm
center feed 2.3 mm
4.0 mm
0.8 mm
4.8 mm
12.0 mm
Figure D.8: Top view of conductor-backed folded slot design with periodic dielectric sub-strate (a=1.2 cm, b=4.8 mm, and εr=10.2) where slot dimension in parenthesesrefers only to the reference slot design on an homogeneous substrate
design, was fabricated using wet etching on a square copper clad RT/duroid homogeneous
substrate with dielectric constant of 2.2, a 127 mm (5 in) edge length, and a thickness of
1.65 mm (65 mil). The reference slot was then center-fed using a simple coaxial line [46].
Note that the reference slot is simply a truncated conductor-backed slot with a homogeneous
dielectric substrate. The periodically-machined dielectric shown in the figure corresponds
to the following folded slot design.
D.2.2 Folded Slot
Similarly to the reference slot, a folded slot of length 18.0 mm and width of 2.0 mm was
fabricated using wet etching on a square copper clad RT/duroid substrate with dielectric
constant of 2.2, a 127 mm (5 in) edge length, but a thickness of only 127 µm (5 mil). A top
view of the conductor-backed folded slot design integrated with periodic dielectric substrate
159
(a=1.2 cm, b=4.8 mm, and εr=10.2) is shown in Figure D.8. The difference in size of the
two slots allows for resonances at equivalent frequencies. In order to design the slot antenna
for optimal performance, the slot is designed to resonate near the center frequency of the
designed substrate band-gap. When the folded slot design is integrated with the substrate,
as shown in Figure D.8, the energy from the slot is coupled into the parallel-plate mode
which is then stored in the dielectric lattice, loading the slot inductively. Consequently, an
off-center feed is used to provide a better impedance match.
The thin substrate was then bonded to a square copper clad RT/duroid substrate with
dielectric constant of 10.2, a 127 mm (5 in) edge length, and a thickness of 2.54 mm (100 mil)
milled as shown in Figure D.8. A small amount of dielectric (≈ 10 mil) was left on the lower
plate in order to provide some stability for mounting the slot plate. Simulations have shown
that the remaining small amounts of dielectric (≈ 10% of the total substrate height) on the
back plate change the center frequency and attenuation per period of the band-gap slightly
(≈ 5%). This observation is similar to the small change in the propagation constant of a
partially-loaded parallel-plate waveguide. For instance, the low frequency approximation
for the propagation constant β in an partially-loaded parallel plate waveguide is [24]
β =
√εrb
a+ εr(b− a)k0 =√εek0 (D.1)
where a is the dielectric sheet thickness, b is the plate separation, and εe is the effective
dielectric constant. For a frequency of 10 GHz, plate separation of 2.54 mm, dielectric sheet
thickness of 0.254 mm, and dielectric constant 10.2, the effective dielectric constant εe is
estimated to be 1.1, corresponding to less than a 5% change the propagation constant.
To provide additional support to mount the slot plate, material at the outer edge of
the substrate was not milled as can also be seen in Figure D.8. HFSS simulations have
shown that this supporting material does not significantly affect the performance of the
slot (see Section D.3). The frequency response of the fabricated folded slot design, with
noticeable resonances at 9.4 GHz and 9.6 GHz, is displayed in Figure D.9. The slots were
designed to radiate at 9.2 GHz. Unfortunately, difficulty in bonding the slot plate to the
perforated substrate may have introduced additional unwanted modes and/or a shifting of
the slot resonance. In the same figure, the finite-element simulation (HFSS) of the folded
slot design finds a resonance slightly below 9.0 GHz. The difference between the simulations
and the experiment is due to the simplifications introduced in the structure analyzed by
HFSS to reduce computer memory and simulation time and the fabrication errors mentioned
previously.
160
Frequency (GHz)
5 6 7 8 9 10 11
|S11
| (dB
)
-25
-20
-15
-10
-5
0
MeasuredSimulated
Figure D.9: Measured and simulated frequency response of conductor-backed slot with pe-riodic dielectric substrate
D.3 Results and Discussion
Simulations of coaxially-fed slots in a metal plate in free space have relatively large
10 dB bandwidths on the order of 10–15%. However, since the slot radiates equally into
each half-space, the front-to-back ratio is 0 dB. If a metal sheet is used as a reflector, the
front-to-back ratio is increased, but unwanted energy is trapped in parallel-plate modes
and radiates away from the slot. Simulations of a finite-sized reflector-backed slot with an
absorbing boundary condition at the edges of the resulting cavity, effectively modeling an
infinite parallel-plate, yield a 10 dB bandwidth of 30%. The large bandwidth is the result of
the slot radiating most of its power into the infinite waveguide. If, however, the infinite sheet
is replaced by a finite one, an undulating field pattern will be seen. This is particularly true
for ground planes that are larger than one-half wavelength [46]. Consequently, the 127 mm
edge length of the sheet in this design, corresponding to about 4 free space wavelengths,
yields significant pattern degradation. Notice the classic interference pattern (7–8 dB “dips”
in the pattern) reported earlier [13, 79] to be caused by the radiation from the edges of the
finite ground can be clearly seen in the pattern for the reference in Figure D.10. Some
suppression technique must be implemented to eliminate this parasitic radiation.
In order to reduce the effects of the finite ground, a number of suppression techniques
are known, including integrating the antenna on a substrate lens [31]. Although useful for
161
-20
-10
0-180
-150
-120
-90
-60
-30
0
30
60
90
120
150
E plane (co-pol)E plane (x-pol)H plane (co-pol)H plane (x-pol)
Figure D.10: Measured normalized E - and H -plane antenna patterns (co- and x-pol) ofreference slot at 9.7 GHz
providing unidirectional radiation patterns by suppressing substrate surface wave formation,
lenses are not necessarily low-profile. Cavity-backed slots are another effective method of
increasing the front-to-back ratio. A simulation of a λ/4 cavity-backed slot [46] yielded a
10 dB bandwidth of 7.5%. Careful design can minimize the necessarily narrower bandwidth
of cavity-backed slots. If vertical integration space is a premium, cavity-backed designs
much like a lens may not be appropriate. Recently, a slot on a synthesized three-dimensional
metallic photonic band-gap crystal was fabricated by composing alternating layers of thin
metallic rods [53]. Unfortunately the response of the slot is very sensitive to the placement
of the slot over the rods. Gains of 2–3 dB were reported with low cross-polarization levels
for specific slot locations and for narrow frequency bands. The bandwidth and frequency
response of the antenna was not reported. Although the three-dimensional electromagnetic
crystal acts as a good reflector, it does not enhance the transmission signal as much as
was expected. As was mentioned earlier, parasitic surface states, which reduce the radiated
power, can exist when the periodicity of the three-dimensional structure is broken.
The field distribution inside a slot-fed finite parallel-plate is similar to that of a slot-fed
metallic cavity with PMC edge boundaries. Reflections from the edges of the parallel-plate
waveguide effectively create an over-moded cavity. This parasitic radiation from the edges
is dependent on the mode that is formed in the cavity. Energy leakage from the substrate
162
through the edges of the cavity manifests itself in unwanted effects such as reduced front-
to-back ratio, reduced efficiency, increased cross-polarization level, and pattern distortion
as shown in the reference antenna pattern of Figure D.10. This effect is significantly re-
duced when the specially designed periodic dielectric substrate is implemented. If the slot
is designed to radiate at a frequency inside the band-gap, the parallel-plate TEM mode
will be trapped in the dielectric lattice. Consequently, very little power is lost to substrate
modes. An effective cavity is formed, since the energy in the parallel-plate mode is re-
flected back to the radiating slot (Figure D.11). The field structure inside the parallel plate
waveguide shown in Figure D.11 was determined from a HFSS simulation implemented
by Chappell [91]. The darker regions in the figure are areas of higher field intensity; the
lighter regions are areas of lower field intensity. As mentioned previously, the supporting
material remaining at the edge of the substrate does not affect the performance of the slot.
The removed material was replaced with a metallic boundary condition and the simulations
were repeated. No significant change in field strength or field structure was observed. The
periodic dielectric substrate has two benefits over traditional metallic cavities. Namely, the
reflection from the cavity boundaries can be controlled – adding more layers of periodic
material around the slot will increase the reflectivity of the boundaries – and the boundary
conditions of the resulting effective cavity are frequency dependent.
Figure D.11: Evanescent field structure inside EBG-backed substrate (left) and propagatingmode in homogeneously-filled parallel-plate (right)
As can be clearly seen in Figure D.12, the slot has a front-to-back ratio of more than
15 dB. Low cross-polarization levels were measured in both the E and H planes. The 10 dB
bandwidth is measured to be 7.5%, lower than that of the single planar slot and similar
to the bandwidth of traditional cavity-backed designs as expected. It is observed that the
antenna patterns are similar to that of a slot in an infinite ground plane; the significant
163
-20
-10
0-180
-150
-120
-90
-60
-30
0
30
60
90
120
150
E plane (co-pol)E plane (x-pol)H plane (co-pol)H plane (x-pol)
Figure D.12: Measured normalized E - and H -plane antenna patterns (co- and x-pol) ofconductor-backed slot with EBG substrate at 9.7 GHz
7–8 dB “dips” in the reference slot pattern are reduced to 1–2 dB in the pattern of the slot
backed by the periodic dielectric (Figure D.13).
-20
-10
0-180
-150
-120
-90
-60
-30
0
30
60
90
120
150
E plane (EMC)E plane (reference)
Figure D.13: Measured normalized E -plane antenna patterns (co-pol) of the reference slotand the slot backed with the periodic dielectric at 9.7 GHz
164
D.3.1 Gain Measurement
A common method for gain measurement is the relative gain measurement technique
or gain-transfer (gain-comparison) method [8]. This approach is based on the substitution
of a test antenna (unknown gain) with an antenna with a known gain (standard gain) and
comparing the corresponding received power levels. Since all of the parameters are fixed
except for the gain of the receiver antenna, it can be easily shown that the Friis Transmission
equation reduces to
Gu =P urP srGs (D.2)
where Gu and Gs are the gains of the unknown and standard antennas, respectively, and
P ru and P rs are the received powers for the antenna under test and the standard gain an-
tenna, respectively. The maximum received powers (E -plane) for the slotted antenna and
for a Narda standard gain horn antenna were measured and the gain of the conductor-
backed antenna mounted over the artificial dielectric substrate antenna was found to be
approximately 3.1±0.2 dB.
The field patterns of the slot antenna mounted on the EBG-backed substrate shown in
the preceding figures are similar to those obtained for an ideal slot. For an ideal, perfectly
matched, one-half wavelength slot mounted in an perfectly conducting infinite plate, the
variation of Eθ as a function of θ can be approximated by
Eθ =cos [(π/2) cos θ]
sin θ(D.3)
and the variation of Eθ as a function of φ is constant.
D.3.2 Directivity Measurement
Directivity is a indication of the directional properties of an antenna. The directivity
can be found by finding the maximum increase in power density in a given direction from
a fixed transmitting antenna, relative to the power density with the same transmit power
distributed equally in all directions [109]. Accurate determinations of the directivity require
that the full three-dimensional antenna pattern be known or measured. Because a full (θ, φ)
pattern cannot be obtained easily for all antennas, a number of different methods have been
proposed to approximate the directivity. The predicted directivity (gain) for an ideal slot
in infinite ground is 1.6 (2.2 dB) [46]. For the ideal slot radiating only into the upper
hemisphere, the directivity (gain) is doubled to approximately 3.2 (5.2 dB) [13].
165
Principal-plane or half-power beamwidth approximation
An approximate value for the directivity D of antenna can be found by dividing the
directivity-beamwidth product DB by the principal-plane beamwidths, HPE and HPH ,
D =DB
HPEHPH. (D.4)
Appropriate values for DB range from 4π(180/π)2 = 41 253 deg2 for a rectangular beam
with no-sidelobe pattern to more realistic values of 30 000 [8] and 26 000 [110, 109] for
practical antenna measurements. Using the latter value of 26 000 as a worst-case scenario
for the directivity-beamwidth product and determining the principal-plane beamwidths of
the folded slot to be 160 and 60, a coarse estimate for the directivity is 2.7 (4.3 dB)
yielding an efficiency of about 80%.
x
y
z
φ
θ
E-plane
H-plane
Figure D.14: Axis and pattern cut definitions of the measurement systems
Pattern integration
Although half-power beamwidth measurements are simple antenna measurements that
provide a coarse estimate for the directivity, more accurate calculations can be performed
by integrating known θ and φ cuts. Assuming the radiation intensity of a given antenna
is separable in θ and φ, an estimate for the total radiated power can be determined by
integrating Eθ and Eφ over the E -plane (θ) shown in Figure D.14 while assuming the φ-
variation is constant. By integrating the pattern over θ and multiplying the result by 2π,
an estimate for the directivity is 2.76 (4.4 dB) yielding an efficiency of around 80%, similar
166
to the efficiency obtained previously. More accurate values of the radiated power density,
and consequently efficiency, can be obtained by introducing additional θ and φ cuts.
In order to verify that the estimated efficiencies calculated for the antenna are reason-
able, HFSS simulations of the finite-size, conductor-backed folded-slot antenna mounted
over the designed artificial dielectric were conducted by Chappell [91] that yield an effi-
ciency of over 90%, which is significantly higher than the estimates determined from the
measurements. However, it should be noted that the simulation assumed no conductor loss
which is one of the dominant loss mechanisms in the system.
D.4 Conclusions
Although relatively thick substrates were used in this chapter to demonstrate the useful-
ness of the conductor-backed folded slot with periodic dielectric substrates, similar results
can be obtained using significantly thinner substrates. This is very desirable for applications
involving packaging, such as layered circuits or vertical integration of components, particu-
larly where ground plane proximity is a concern. In addition, the effective cavity resulting
from the lattice is frequency dependent and tunable. Above and below the band-gap, the
periodic structure allows energy to propagate through virtually unimpeded. In the center of
the gap, the structure is virtually impassible to electromagnetic propagation. Near the band
edges, proper design could further increase the bandwidth of this antenna by decreasing the
attenuation per period or by allowing specific modes to propagate unattenuated.
167
APPENDIX E
Review Of The Method Of Moments
The use of the methods of moments [34] to numerically solve various electromagnetic
problems has been used effectively since the mid-1960’s. It has since become one of the
standard techniques used by computational electromagnetic (CEM) theorists because of its
accuracy and ease of implementation.
Consider the following nonhomogeneous equation
Lu = f (E.1)
where L is a linear operator, f is known vector quantity , and u is unknown vector quantity
to be determined. In electromagnetics, L is often a linear intergro-differential operator, f is
the known excitation, and u is the unknown field or current. We begin by expanding u in
a linear combination of basis (or expansion) functions
u =∑n
anun (E.2)
where an are the coefficients of the basis un. Finite computational resources require the
sum in (E.2) to be finite. Thus, the method of moments, although exact in theory, is an
approximate technique in practice. Substituting (E.2) into (E.1) one obtains
∑n
anLun = f . (E.3)
To determine the coefficients of the unknown basis, (E.3) is tested by performing an
inner product of the linear combination of basis functions with N weighting (or testing)
functions
∑n
an 〈wm, Lun〉 = 〈wm, f〉 m = 1, 2, . . . , N (E.4)
168
where the inner product 〈a,b〉 can be defined as
〈a,b〉 =∫
a · b∗ dx (E.5)
where x is a multidimensional variable and ∗ denotes the complex conjugate. (E.4) can be
written in matrix form as
[Lmn
] [an
]=[fm
], (E.6)
where
[Lmn
]=
〈w1, Lu1〉 〈w1, Lu2〉 · · · 〈w1, LuN 〉〈w2, Lu1〉 〈w2, Lu2〉 · · · 〈w2, LuN 〉
......
. . ....
〈wN , Lu1〉 〈wN , Lu2〉 · · · 〈wN , LuN 〉
, (E.7a)
[an
]=
a1
a2
...
aN
, (E.7b)
and
[fm
]=
〈w1, f〉〈w2, f〉
...
〈wN , f〉
. (E.7c)
In electromagnetics, [Lmn] is often represented as an impedance matrix [Zmn] relating the
excitation to the unknown.
If [Lmn]−1 exists, the solution for the coefficients an can be found from
[an
]=[Lmn
]−1 [fm
], (E.8)
using a number of well-known linear solution techniques.
169
BIBLIOGRAPHY
170
BIBLIOGRAPHY
[1] Journal of the Optical Society of America B, vol. 10, Feb 1993. Special issue onphotonic band-gap structures.
[2] Journal of Modern Optics, vol. 41, Feb 1994. Special issue on photonic band-gapstructures.
[3] IEEE Transactions on Microwave Theory and Techniques, vol. 47, Nov 1999. Mini-special issue on electromagnetic crystal structures, design, synthesis, and applications.
[4] Journal of Lightwave Technology, vol. 17, Nov 1999. Special section on electromagneticcrystal structures, design, synthesis, and applications.
[5] Electromagnetics, vol. 41, Feb 1999. Special issue on theory and applications of pho-tonic band-gap structures.
[6] N. G. Alexopoulos, “Integrated-circuit structures on anisotropic substrates,” IEEETransactions on Microwave Theory and Techniques, vol. MTT-33, no. 10, pp. 847–881, Oct 1985.
[7] I. J. Bahl, P. Bhartia, and S. S. Stuchly, “Design of microstrip antennas covered witha dielectric layer,” IEEE Transactions on Antennas and Propagation, vol. AP-30, no.2, pp. 314–317, Mar 1982.
[8] C. A. Balanis, Antenna Theory: Analysis and Design, Harper & Row Publishers, NewYork, NY, 1982.
[9] C. A. Balanis, Advanced Engineering Electromagnetics, John Wiley & Sonce, Inc.,New York, NY, 1989.
[10] H. Bertoni, L.-H. Cheo, and T. Tamir, “Frequency-selective reflection and transmis-sion by a periodic dielectric layer,” IEEE Transactions on Antennas and Propagation,vol. AP-37, no. 1, pp. 78–83, Jan 1989.
[11] P. Bhartia, K. V. S. Rao, and R. S. Tomar, Millimeter-wave Microstrip and PrintedCircuit Antennas, Artech House, Boston, MA, 1991.
[12] A. K. Bhattacharyya, “Analysis of multilayer infinite periodic array structures withdifferent periodicities and axes orientation,” IEEE Transactions on Antennas andPropagation, vol. AP-48, no. 3, pp. 357–369, Mar 2000.
[13] H. G. Booker, “Slot aerials and their relation to complementary wire aerials,” Journalof the Institution of Electrical Engineers, vol. 93, no. 4, , 1946. IIIA.
171
[14] F. Bouttout, F. Benabdelaziz, A. Benghalia, D. Khedrouche, and T. Fortaki, “Uni-axially anisotropic substrate effects on resonance of rectangular microstrip patch an-tenna,” Electronics Letters, vol. 35, no. 4, pp. 255–256, Feb 1999.
[15] C. Brezinski, “Some new convergence acceleration methods,” Mathematics of Com-puting, vol. 39, pp. 133–145, 1982.
[16] L. Brillouin, Wave Propagation in Periodic Structures, McGraw–Hill, New York, NY,1946.
[17] E. R. Brown and O. B. McMahon, “High zenithal directivity from a dipole antennaon a photonic crystal,” Applied Physics Letters, vol. 68, no. 9, pp. 1300–1302, Feb1996.
[18] E. R. Brown, C. D. Parker, and E. Yablonovitch, “Radiation properties of a planarantenna on a photonic-crystal substrate,” Journal of the Optical Society of AmericaB, vol. 10, no. 2, pp. 404–407, Feb 1993.
[19] Z. Cai and J. Bornemann, “Rigorous analysis of radiation properties of lossy patchresonantors on complex anisotropic media and lossy ground metallization,” IEEETransactions on Antennas and Propagation, vol. AP-42, no. 10, pp. 1443–1446, Oct1994.
[20] C. H. Chan, “Analysis of frequency selective surfaces,” in Frequency Selective Surfaceand Grid Array, T. K. Wu, editor, chapter 2, pp. 27–85, John Wiley & Sons, Inc.,New York, NY, 1985.
[21] S. D. Cheng, R. Biswas, E. Ozbay, S. McCalmont, G. Tuttle, and K.-M. Ho, “Opti-mized dipole antennas on photonic band gap crystals,” Applied Physics Letters, vol.67, no. 23, pp. 3399–3401, Dec 1995.
[22] D. N. Chigrin, A. V. Lavrinenko, D. A. Yarotsky, and S. V. Gaponenko, “Observationof total omnidirectional reflection from a one-dimensional dielectric lattice,” AppliedPhysics A, vol. 68, pp. 25—28, 1999.
[23] R. E. Collin, “A simple artificial anisotropic dielectric medium,” IRE Transactionson Microwave Theory and Techniques, pp. 206–209, 1958.
[24] R. E. Collin, Foundations for Microwave Engineering, McGraw–Hill, Inc., New York,NY, 1992.
[25] S. R. Coriell and J. L. Jackson, “Bounds on transport coefficients of two-phase mate-rials,” Journal of Applied Physics, vol. 39, no. 10, pp. 4733–4736, September 1968.
[26] J. P. Dowling and C. M. Bowden, “Anomalous index of refraction in photonic bandgapmaterials,” Journal of Modern Optics, vol. 41, no. 2, pp. 345–351, 1994.
[27] T. F. Eibert, J. L. Volakis, D. R. Wilton, and D. R. Jackson, “Hybrid FE/BI model-ing of 3-D doubly periodic structures utilizing triangular prsimatic elements and anMPIE formulation accelerated by the Ewald transformation,” IEEE Transactions onAntennas and Propagation, vol. AP-47, no. 5, pp. 843—850, May 1999.
172
[28] R. S. Elliott, Antenna Theory and Design, Prentice–Hall, Inc., Englewood Cliffs, NJ,1981.
[29] T. Ellis, J. D. Shumpert, L. P. B. Katehi, and G. M. Rebeiz, “Do two dimensional peri-odic structures exhibit photonic bandgap properties?,” in 1998 URSI North AmericanRadio Science Meeting Digest, p. 120, Jun 1998.
[30] P. P. Ewald, “Die Berechnung optisher und elektrostatischer Gitterpotentiale,” An-nalen der Physik, vol. 64, no. 3, pp. 253–286, 1921.
[31] D. F. Filipovic, S. S. Gearhart, and G. M. Rebeiz, “Double-slot antennas on ex-tended hemi-spherical and elliptical silicon dielectric lenses,” IEEE Transactions onMicrowave Theory and Techniques, vol. MTT-41, no. 10, pp. 1738–1749, Oct 1993.
[32] S. D. Gedney, J. F. Lee, and R. Mittra, “A combined FEM/MoM approach to analyzethe plane wave diffraction by arbitrary gratings,” IEEE Transactions on MicrowaveTheory and Techniques, vol. MTT-40, no. 2, pp. 363–370, Feb 1992.
[33] R. F. Harrington, Time-Harmonic Electromagnetic Fields, McGraw–Hill, New York,NY, 1961.
[34] R. F. Harrington, Field Computation by Moment Methods, Kreiger Publishing Com-pany, Malabar, FL, 1993.
[35] Z. Hashin and S. Shtrikman, “A variational approach to the theory of effective mag-netic permeability of multiphase materials,” Journal of Applied Physics, vol. 33, pp.3125–3131, 1962.
[36] K. M. Ho, C. T. Chan, and C. M. Soukoulis, “Existence of a photonic gap in periodicdielectric structures,” Physical Review Letters, vol. 65, no. 25, pp. 3152–3155, Dec1990.
[37] D. R. Jackson and N. G. Alexopoulos, “Simple approximate formulas for input resis-tance, bandwidth, and efficiency of a resonant rectangular patch,” IEEE Transactionson Antennas and Propagation, vol. AP-39, no. 3, pp. 407–410, Mar 1991.
[38] J. D. Joannopoulos, R. D. Meade, and J. N. Winn, Photonic Crystals: Molding theFlow of Light, Princeton University Press, Princeton, NJ, 1995.
[39] K. E. Jordan, G. R. Richter, and P. Sheng, “An efficient numerical evaluation ofthe Green’s function for the Helmholtz operator on periodic structures,” Journal ofComputational Physics, vol. 63, pp. 222–235, 1986.
[40] R. E. Jorgenson and R. Mittra, “Efficient calculation of the free-space Green’s func-tion,” IEEE Transactions on Antennas and Propagation, vol. AP-38, no. 5, pp. 633–642, May 1990.
[41] M. P. Kesler, J. G. Maloney, B. L. Shirley, and G. S. Smith, “Antenna design withthe use of photonic band-gap materials as all-dielectric planar reflectors,” Microwaveand Optical Technology Letters, vol. 11, no. 4, pp. 169–174, Mar 1996.
[42] N. Kinayman and M. I. Aksum, “Comparative study of acceleration techniques forintegrals and series in electromagnetic problems,” Radio Science, vol. 30, no. 6, pp.1713–1722, Nov–Dec 1995.
173
[43] R. L. Kipp and C. H. Chan, “A numerically efficient technique for the method ofmoments solution for planar periodic structures in layered media,” IEEE Transactionson Microwave Theory and Techniques, vol. MTT-42, no. 4, pp. 635–643, Apr 1994.
[44] C. Kittel, Introduction to Solid State Physics, John Wiley & Sons, Inc., New York,NY, seventh edition, 1996.
[45] J. A. Kong, Electromagnetic Wave Theory, John Wiley & Sons, Inc., New York, NY,1990.
[46] J. D. Kraus, Antennas, McGraw–Hill Publishing, New York, NY, 1950.
[47] E. F. Kuester and C. L. Holloway, “Comparison of approximations for effective pa-rameters of artificial dielectrics,” IEEE Transactions on Microwave Theory and Tech-niques, vol. MTT-38, no. 11, pp. 1752–1755, Nov 1990.
[48] E. F. Kuester and C. L. Holloway, “A low-frequency model for wedge or pyramidabsorber arrays — I: theory,” IEEE Transactions on Electromagnetic Compatability,vol. EMC-36, no. 4, pp. 300–306, Nov 1994.
[49] C. A. Kyriazidou, H. F. Contopanagos, W. M. Merrill, and N. G. Alexopoulos, “Ar-tificial versus natural crystals: effective wave impedance of printed photonic bandgapmaterials,” IEEE Transactions on Antennas and Propagation, vol. AP-48, no. 1, pp.95–104, Jan 2000.
[50] P. Lalanne, “Effective medium theory applied to photonic crystals composed of cubicor square cylinders,” Applied Optics, vol. 35, no. 27, pp. 5369–5380, Sep 1996.
[51] R. Lampe, P. Klock, and P. Mayes, “Integral transforms useful for the accelerated sum-mation of periodic, free-space Green’s functions,” IEEE Transactions on MicrowaveTheory and Techniques, vol. MTT-33, no. 8, pp. 734–736, Aug 1985.
[52] K. M. Leung and Y. F. Liu, “Full vector calculation of photonic band structures inface-centered-cubic dielectric media,” Physical Review Letters, vol. 65, no. 21, pp.2646–2649, Nov 1990.
[53] W. Y. Leung, R. Biswas, S.-D. Cheng, M. M. Sigalas, J. S. McCalmont, G. Tuttle,and K.-M. Ho, “Slot antennas on photonic band gap crystals,” IEEE Transactions onAntennas and Propagation, vol. AP-45, no. 10, pp. 1569–1570, Oct 1997.
[54] D. Levin, “Development of nonlinear transformations for improving convergence ofsequences,” International Journal of Computational Mathematics, vol. 3, no. B, pp.371–388, 1973.
[55] L. R. Lewis and A. Hessel, “Propagation constants of periodic arrays of dielectricslabs,” IEEE Transactions on Microwave Theory and Techniques, vol. MTT-19, pp.276–286, Mar 1971.
[56] D. E. Livesay and K.-M. Chen, “Electromagnetic field induced inside arbitrary shapedbiological bodies,” IEEE Transactions on Microwave Theory and Techniques, vol.MTT-22, no. 12, pp. 1273–1280, Dec 1974.
[57] J. Maddox, “Do photonic band-gaps really exist?,” Nature, vol. 348, pp. 481, 1990.
174
[58] A. A. Maradudin and A. R. McGurn, “Photonic band structure of a truncated, two-dimensional, periodic dielectric medium,” Journal of the Optical Society of AmericaB, vol. 10, no. 2, pp. 307–313, Feb 1993.
[59] A. A. Maradudin and A. R. McGurn, “Out-of-plane propagation of electromagneticwaves in a two-dimensional periodic dielectric medium,” Journal of Modern Optics,vol. 41, no. 2, pp. 275–284, 1994.
[60] S. L. McCall, P. M. Platzman, R. Dalichaouch, D. Smith, and S. Schultz, “Microwavepropagation in two-dimensional dielectric lattice,” Physical Review Letters, vol. 67,no. 17, pp. 2017–2020, Oct 1991.
[61] D. T. McGrath and V. P. Pyati, “Periodic structure analysis using a hybrid finiteelement method,” Radio Science, vol. 31, no. 5, pp. 1173–1179, Sep-Oct 1996.
[62] A. R. McGurn and A. A. Maradudin, “Photonic band structures of two- and three-dimensional periodic metal or semiconductor arrays,” Physics Review B, vol. 48, no.23, pp. 17576–17579, Dec 1993.
[63] R. D. Meade, O. Alerhand, and J. D. Joannopoulos, Handbook of Photonic Band-GapMaterials, JAMTex, Belmont, MA, 1993.
[64] R. D. Meade, K. D. Brommer, A. M. Rappe, and J. D. Joannopoulos, “Existence ofa photonic band gap in two dimensions,” Applied Physics Letters, vol. 61, no. 27, pp.495–497, Jul 1992.
[65] E. Michielssen, Electromagnetic Plane Wave Scattering from Periodically LoadedCylinders and Doubly Periodic Slabs, PhD dissertation, University of Illinois, Urbana-Champaign, 1992.
[66] R. Mittra, C. H. Chan, and T. Cwik, “Techniques for analyzing frequency selectivesurfaces – a review,” Proceedings of the IEEE, vol. 76, no. 12, pp. 1593–1614, Dec1988.
[67] B. A. Munk and G. A. Burrell, “Plane wave expansion for arrays of arbitrarily ori-ented piecewise linear elements and its application in determining the impedance ofa single linear antenna in a lossy half-space,” IEEE Transactions on Antennas andPropagation, vol. AP-27, no. 3, pp. 331–343, May 1979.
[68] R. M. Nelson, D. A. Rogers, and A. G. D’Assuncao, “Resonant frequency of a rect-angular microstrip patch on several uniaxial substrates,” IEEE Transactions on An-tennas and Propagation, vol. 38, no. 7, pp. 973–981, Jul 1990.
[69] A. A. Oliner, “Periodic structures and photonic band-gap terminology: historicalperspectives,” in 29th European Microwave Conference, Oct 1999.
[70] A. Papoulis, The Fourier Integral and Its Applications, McGraw–Hill, New York, NY,1962.
[71] M.-J. Park, J. Park, and S. Nam, “Efficient calculation of the Green’s function forthe rectangular cavity,” IEEE Microwave and Guided Wave Letters, vol. 8, no. 3, pp.1958–1962, Mar 1998.
175
[72] S. T. Peng, T. Tamir, and H. L. Bertoni, “Theory of periodic dielectric waveguides,”IEEE Transactions on Microwave Theory and Techniques, vol. MTT-23, no. 1, pp.123–133, Jan 1975.
[73] M. Plihal, A. Shambrook, A. A. Maradadin, and P. Sheng, “Two-dimensional photonicband structures,” Optics Communications, vol. 80, no. 3/4, pp. 199–204, Jan 1991.
[74] D. M. Pozar, “Radiation and scattering from a microstrip patch antenna on a uniaxialsubstrate,” IEEE Transactions on Antennas and Propagation, vol. AP-35, no. 6, pp.613–621, Jun 1987.
[75] D. M. Pozar, “Microstrip antennas,” Proceedings of the IEEE, vol. 80, no. 1, pp.79–91, Jan 1992.
[76] Y. Rahmat-Samii, “On the question of computation of the dyadic Green’s functionat the source region in waveguide cavities,” IEEE Transactions on Microwave Theoryand Techniques, vol. MTT-23, no. 9, pp. 762–765, Sept 1975.
[77] R. Remski, “Modeling and optimizing photonic bandgap (PBG) structures,” in Ansoft“No Boundaries” Workshop, Baltimore, MD, October 2000.
[78] P. A. Rizzi, Microwave Engineering, Prentice Hall, Englewood Cliffs, NJ, 1988.
[79] S. V. Robertson, L. P. B. Katehi, and G. M. Rebeiz, “A planar quasi-optical mixer us-ing a folded-slot antenna,” IEEE Transactions on Microwave Theory and Techniques,vol. MTT-43, no. 4, pp. 896–898, Apr 1995.
[80] P. Russell, “Photonic band-gaps,” Physics World, vol. 5, no. 8, pp. 37–42, 1 Aug 1992.
[81] D. B. Rutledge, D. P. Neikirk, and D. P. Kasilingham, “Integrated-circuit antennas,”in Millimeter Components and Techniques, Part II, K. J. Button, editor, vol. 10 ofInfrared and Millimeter Waves, chapter 1, pp. 1–90, Academic Press, Orlando, FL,1983.
[82] S. M. Rytov, “Electromagnetic properties of a finely stratified medium,” SovietPhysics Journal of Experimental and Theoretical Physics of the Academy of Scienceof the USSR, vol. 2, pp. 466–475, 1956.
[83] M. N. O. Sadiku, Numerical Techniques in Electromagnetics, CRC Press, Boca Raton,FL, 1992.
[84] K. Sarabandi, Electromagnetic Scattering from Vegetation Canopies, PhD disserta-tion, University of Michigan, 1989.
[85] K. Sarabandi, “Simulation of a periodic dielectric corrugation with an equivalentanisotropic layer,” International Journal of Infrared and Millimeter Waves, vol. 11,no. 11, pp. 1303–1321, Nov 1990.
[86] K. Sarabandi and P. F. Polatin, “Electromagnetic scattering from two adjacent ob-jects,” IEEE Transactions on Antennas and Propagation, vol. AP-42, no. 4, pp. 510–517, Apr 1994.
176
[87] K. Sarabandi and F. T. Ulaby, “High frequency scattering from corrugated stratifiedcylinders,” IEEE Transactions on Antennas and Propagation, vol. AP-39, no. 4, pp.512–520, Apr 1991.
[88] M. Scalora, M. J. Bloemer, J. P. Dowling, C. M. Bowden, and A. S. Manka, “Trans-parent, metallo-dielectric, one-dimensional, photonic band-gap structures,” Journalof Applied Physics, vol. 83, no. 5, pp. 2377–2383, Mar 1998.
[89] D. H. Schaubert, “Microstrip antennas,” Electromagnetics, vol. 12, pp. 381–401, 1992.
[90] D. Shanks, “Non-linear transformations of divergent and slowly convergent se-quences,” Journal of Mathematical Physics, vol. 34, pp. 1–42, 1955.
[91] J. D. Shumpert, W. J. Chappell, and L. P. B. Katehi, “Parallel-plate mode reductionin conductor-backed slots using electromagnetic band-gap substrates,” IEEE Trans-actions on Microwave Theory and Techniques, vol. MTT-47, no. 11, pp. 2099–2105,Nov 1999.
[92] J. D. Shumpert, T. Ellis, G. M. Rebeiz, and L. P. B. Katehi, “Microwave andmillimeter-wave propagation through photonic band-gap structures,” in 1997 URSINorth American Radio Sciece Meeting Digest, p. 678, Jul 1997.
[93] J. D. Shumpert, T. Ellis, G. M. Rebeiz, and L. P. B. Katehi, “Microwave andmillimeter-wave propagation in photonic band-gap structures,” Radiation Lab Re-port RL-953, The University of Michigan, Oct 1997.
[94] J. D. Shumpert, T. Ellis, G. M. Rebeiz, and L. P. B. Katehi, “Two-dimensional band-gap formation in finite substrates,” in 1998 URSI North American Radio ScienceMeeting Digest, p. 123, Jun 1998.
[95] M. M. Sigalas, R. Biswas, and K.-M. Ho, “Theoretical study of dipole antennas onphotonic band-gap materials,” Microwave and Optical Technology Letters, vol. 13, no.4, pp. 205–209, Nov 1996.
[96] M. M. Sigalas, R. Biswas, K. M. Ho, and C. M. Soukoulis, “Theoretical investigation ofoff-plane propagation of electromagnetic waves in two-dimensional photonic crystals,”Physical Review B, vol. 58, no. 11, pp. 6791–6794, Sept 1998.
[97] M. M. Sigalas, R. Biswas, Q. Li, D. Crouch, W. Leung, R. Jacobs-Woodbury,B. Laogh, S. Nielsen, S. McCalmont, G. Tuttle, and K.-M. Ho, “Dipole antennason photonic band-gap crystals: experiment and simulation,” Microwave and OpticalTechnology Letters, vol. 15, no. 3, pp. 153–158, Jun 1997.
[98] J. Singh, Quantum Mechanics: Fundamentals and Applications to Technology, JohnWiley & Sons, Inc., New York, NY, 1997.
[99] S. Singh, W. F. Richards, J. R. Zinecker, and D. R. Wilton, “Accelerating the conver-gence of series representing the free space periodic Green’s function,” IEEE Transac-tions on Antennas and Propagation, vol. AP-38, no. 12, pp. 1958–1962, Dec 1990.
[100] S. Singh and R. Singh, “Application of transforms to accelerate the summation ofperiodic free-space Green’s functions,” IEEE Transactions on Microwave Theory andTechniques, vol. MTT-38, no. 11, pp. 1746–1748, Nov 1990.
177
[101] S. Singh and R. Singh, “Efficient computation of the free-space periodic Green’s func-tion,” IEEE Transactions on Microwave Theory and Techniques, vol. MTT-39, no. 7,pp. 1226–1229, Jul 1991.
[102] S. Singh and R. Singh, “On the use of ρ algorithm in series acceleration,” IEEETransactions on Antennas and Propagation, vol. AP-39, no. 10, pp. 1514–1517, Oct1991.
[103] S. Singh and R. Singh, “On the use of Shanks’ transform to accelerate the sum-mation of slowly converging series,” IEEE Transactions on Microwave Theory andTechniques, vol. MTT-39, no. 3, pp. 608–610, Mar 1991.
[104] S. Singh and R. Singh, “A convergence acceleration procedure for computing slowlyconverging series,” IEEE Transactions on Microwave Theory and Techniques, vol.MTT-40, no. 1, pp. 168–171, Jan 1992.
[105] S. Singh and R. Singh, “On the use of Chebyshev-Toeplitz algorithm in acceleratingthe numerical convergence of infinite series,” IEEE Transactions on Microwave Theoryand Techniques, vol. MTT-40, no. 1, pp. 171–173, Jan 1992.
[106] S. Singh and R. Singh, “On the use of Levin’s t-transform in accelerating the summa-tion of series representing the free-space Green’s functions,” IEEE Transactions onMicrowave Theory and Techniques, vol. MTT-41, no. 5, pp. 884–886, May 1993.
[107] D. R. Smith, R. Dalichaouch, N. Kroll, S. Schultz, S. L. McCall, and P. M. Platzman,“Photonic band structure and defects in one and two dimensions,” Journal of theOptical Society of America B, vol. 10, no. 2, pp. 314–321, Feb 1993.
[108] J. A. Stratton, Electromagnetic Theory, McGraw–Hill, New York, NY, 1941.
[109] W. L. Stutzman, “Estimating directivity and gain of antennas,” IEEE Antennas andPropagation Magazine, vol. 40, no. 4, pp. 7–10, Aug 1998.
[110] W. L. Stutzman and G. A. Thiele, Antenna Theory and Design, John Wiley & Sons,Inc., New York, NY, second edition, 1998.
[111] C.-T. Tai, Dyadic Green Functions in Electromagnetic Theory, IEEE Press, New York,NY, second edition, 1994.
[112] A. Tavakoli, K. Sarabandi, and F. T. Ulaby, “Horizontal propagation through periodicvegetation canopies,” IEEE Transactions on Antennas and Propagation, vol. AP-39,no. 7, pp. 1014–1023, Jul 1991.
[113] W.-J. Tsay, Radiation and Scattering from Periodic Geometries in InhomogeneousMedia, PhD dissertation, University of Massachusetts Amherst, 1995.
[114] W.-J. Tsay and D. M. Pozar, “Radiation and scattering from infinite periodic printedantennas with inhomogeneous media,” IEEE Transactions on Antennas and Propa-gation, vol. AP-46, no. 11, pp. 1641–1650, Nov 1998.
[115] B. Veidt, “Selecting 3-D electromagnetic software,” Microwave Journal, vol. 41, no.9, pp. 126–137, Sept 1998.
178
[116] P. R. Villeneuve and M. Piche, “Photonic band gaps in two-dimensional square andhexagonal lattices,” Physical Review B, vol. 46, no. 8, pp. 4969–4972, Aug 1992.
[117] P. R. Villeneuve and M. Piche, “Photonic band gaps in two-dimensional square lat-tices: square and circular rods,” Physical Review B, vol. 46, no. 8, pp. 4973–4975,Aug 1992.
[118] J. J. H. Wang, “Analysis of a three-dimensional arbitrary shaped dielectric or biolog-ical body inside a rectangular waveguide,” IEEE Transactions on Microwave Theoryand Techniques, vol. MTT-26, no. 7, pp. 457–462, Aug 1978.
[119] J. J. H. Wang, Generalized Moment Methods in Electromagnetics: Formulation andComputer Solution of Integral Equations, John Wiley & Sons, Inc., New York, NY,1991.
[120] E. T. Whittaker and G. N. Watson, A Course of Modern Analysis, Cambridge Uni-versity Press, Cambridge, MA, fourth edition, 1927.
[121] J. Wimp, “Toeplitz arrays, linear sequence transformations, and orthogonal polyno-mials,” Numerical Mathematics, vol. 23, pp. 1–17, 1974.
[122] J. N. Winn, R. D. Meade, and J. D. Joannolpoulos, “Two-dimensional photonic band-gap materials,” Journal of Modern Optics, vol. 41, no. 2, pp. 257–273, 1994.
[123] K.-L. Wong, J.-S. Row, C.-W. Kuo, and K.-C. Huang, “Resonance of a rectangularmicrostrip patch on a uniaxial substrate,” IEEE Transactions on Microwave Theoryand Techniques, vol. MTT-41, no. 4, pp. 698–701, Apr 1993.
[124] P. Wynn, “On a device for computing the em(sn) transformation,” MathematicalTables and Other Aids for Computation, vol. 10, pp. 91–96, 1956.
[125] E. Yablonovitch, “Inhibited spontaneous emission in solid-state physics and electron-ics,” Physical Review Letters, vol. 58, no. 2, pp. 2059–2062, 1987.
[126] E. Yablonovitch and T. J. Gmitter, “Photonic band structure: the face-centered-cubiccase,” Physical Review Letters, vol. 63, no. 18, pp. 1950–1953, 1989.
[127] E. Yablonovitch, T. J. Gmitter, and K. M. Leung, “Photonic band structure: the face-centered-cubic case employing non-spherical atoms,” Physical Review Letters, vol. 67,no. 17, pp. 2295–2298, Oct 1991.
[128] C.-F. Yang, W. D. Burnside, and R. C. Rudduck, “A doubly periodic moment methodsolution for the analysis and design of an absorber covered wall,” IEEE Transactionson Antennas and Propagation, vol. AP-41, no. 5, pp. 600–609, May 1993.
[129] H.-Y. Yang, “Finite difference analysis of 2-D photonic crystals,” IEEE Transactionson Microwave Theory and Techniques, vol. MTT-44, no. 12, pp. 2688–2695, Dec 1996.
[130] H.-Y. Yang, “Characteristics of guided and leaky waves on multilayer thin-film struc-tures with planar material gratings,” IEEE Transactions on Microwave Theory andTechniques, vol. MTT-45, no. 3, pp. 428–435, Mar 1997.
[131] H.-Y. Yang and J. A. Castaneda, “Infinite phased arrays of microstrip antennas ongeneralized anisotropic substrates,” Electromagnetics, vol. 11, pp. 107–124, 1991.
179
[132] H.-Y. Yang, J. A. Casteneda, and N. G. Alexopolous, “An integral equation analy-sis of an infinite array of rectangular dielectric waveguides,” IEEE Transactions onMicrowave Theory and Techniques, vol. MTT-38, no. 7, pp. 873–879, Jul 1990.
[133] H.-Y. D. Yang, “Theory of antenna radiation from photonic band-gap materials,”Electromagnetics, vol. 19, no. 3, pp. 255–276, May-June 1999.
[134] H.-Y. D. Yang, “Theory of microstrip lines on artificial periodic substrates,” IEEETransactions on Microwave Theory and Techniques, vol. MTT-47, no. 5, pp. 629–635,May 1999.
[135] H.-Y. D. Yang, N. G. Alexopolous, and E. Yablonovitch, “Photonic band-gap ma-terials for high-gain printed circuit antennas,” IEEE Transactions on Antennas andPropagation, vol. AP-45, no. 1, pp. 185–187, Jan 1997.
[136] H.-Y. D. Yang, R. Diaz, and N. G. Alexopolous, “Reflection and transmission of wavesfrom multilayer structures with planar-implanted periodic material blocks,” Journalof the Optical Society of America B, vol. 14, no. 10, pp. 2513–2521, Oct 1997.
[137] A. Yariv and P. Yeh, Optical Waves in Crystals, John Wiley & Sons, Inc., New York,NY, 1984.
[138] Z. Zhang and S. Satpathy, “Electromagnetic wave propagation in periodic structures:bloch wave solution of Maxwell’s equations,” Physical Review Letters, vol. 65, no. 21,pp. 2650–2653, Nov 1990.
[139] IMSL Fortran 90 MP Numerical Libraries Version 3.0.
180